init commit

2019-04-14 17:34:58 +08:00 · 2019-04-14 17:34:58 +08:00 · 6b8f506591
commit 6b8f506591
parent f8358de8f8
25 changed files with 41706 additions and 0 deletions
--- a/.gitignore
+++ b/.gitignore
@ -0,0 +1,5 @@
+*.pyc
+checkpoints
+*.ckpt
+events.out.*
+.idea
--- a/datasets/init.py
+++ b/datasets/init.py
@ -0,0 +1,7 @@
+from .kitti_dataset import KITTIDataset
+from .sceneflow_dataset import SceneFlowDatset
+
+__datasets__ = {
+    "sceneflow": SceneFlowDatset,
+    "kitti": KITTIDataset
+}
--- a/datasets/data_io.py
+++ b/datasets/data_io.py
@ -0,0 +1,58 @@
+import numpy as np
+import re
+import torchvision.transforms as transforms
+
+
+def get_transform():
+    mean = [0.485, 0.456, 0.406]
+    std = [0.229, 0.224, 0.225]
+
+    return transforms.Compose([
+        transforms.ToTensor(),
+        transforms.Normalize(mean=mean, std=std),
+    ])
+
+
+# read all lines in a file
+def read_all_lines(filename):
+    with open(filename) as f:
+        lines = [line.rstrip() for line in f.readlines()]
+    return lines
+
+
+# read an .pfm file into numpy array, used to load SceneFlow disparity files
+def pfm_imread(filename):
+    file = open(filename, 'rb')
+    color = None
+    width = None
+    height = None
+    scale = None
+    endian = None
+
+    header = file.readline().decode('utf-8').rstrip()
+    if header == 'PF':
+        color = True
+    elif header == 'Pf':
+        color = False
+    else:
+        raise Exception('Not a PFM file.')
+
+    dim_match = re.match(r'^(\d+)\s(\d+)\s$', file.readline().decode('utf-8'))
+    if dim_match:
+        width, height = map(int, dim_match.groups())
+    else:
+        raise Exception('Malformed PFM header.')
+
+    scale = float(file.readline().rstrip())
+    if scale < 0:  # little-endian
+        endian = '<'
+        scale = -scale
+    else:
+        endian = '>'  # big-endian
+
+    data = np.fromfile(file, endian + 'f')
+    shape = (height, width, 3) if color else (height, width)
+
+    data = np.reshape(data, shape)
+    data = np.flipud(data)
+    return data, scale
--- a/datasets/kitti_dataset.py
+++ b/datasets/kitti_dataset.py
@ -0,0 +1,99 @@
+import os
+import random
+from torch.utils.data import Dataset
+from PIL import Image
+import numpy as np
+from datasets.data_io import get_transform, read_all_lines
+
+
+class KITTIDataset(Dataset):
+    def __init__(self, datapath, list_filename, training):
+        self.datapath = datapath
+        self.left_filenames, self.right_filenames, self.disp_filenames = self.load_path(list_filename)
+        self.training = training
+        if self.training:
+            assert self.disp_filenames is not None
+
+    def load_path(self, list_filename):
+        lines = read_all_lines(list_filename)
+        splits = [line.split() for line in lines]
+        left_images = [x[0] for x in splits]
+        right_images = [x[1] for x in splits]
+        if len(splits[0]) == 2:  # ground truth not available
+            return left_images, right_images, None
+        else:
+            disp_images = [x[2] for x in splits]
+            return left_images, right_images, disp_images
+
+    def load_image(self, filename):
+        return Image.open(filename).convert('RGB')
+
+    def load_disp(self, filename):
+        data = Image.open(filename)
+        data = np.array(data, dtype=np.float32) / 256.
+        return data
+
+    def __len__(self):
+        return len(self.left_filenames)
+
+    def __getitem__(self, index):
+        left_img = self.load_image(os.path.join(self.datapath, self.left_filenames[index]))
+        right_img = self.load_image(os.path.join(self.datapath, self.right_filenames[index]))
+
+        if self.disp_filenames:  # has disparity ground truth
+            disparity = self.load_disp(os.path.join(self.datapath, self.disp_filenames[index]))
+        else:
+            disparity = None
+
+        if self.training:
+            w, h = left_img.size
+            crop_w, crop_h = 512, 256
+
+            x1 = random.randint(0, w - crop_w)
+            y1 = random.randint(0, h - crop_h)
+
+            # random crop
+            left_img = left_img.crop((x1, y1, x1 + crop_w, y1 + crop_h))
+            right_img = right_img.crop((x1, y1, x1 + crop_w, y1 + crop_h))
+            disparity = disparity[y1:y1 + crop_h, x1:x1 + crop_w]
+
+            # to tensor, normalize
+            processed = get_transform()
+            left_img = processed(left_img)
+            right_img = processed(right_img)
+
+            return {"left": left_img,
+                    "right": right_img,
+                    "disparity": disparity}
+        else:
+            w, h = left_img.size
+
+            # normalize
+            processed = get_transform()
+            left_img = processed(left_img).numpy()
+            right_img = processed(right_img).numpy()
+
+            # pad to size 1248x384
+            top_pad = 384 - h
+            right_pad = 1248 - w
+            assert top_pad > 0 and right_pad > 0
+            # pad images
+            left_img = np.lib.pad(left_img, ((0, 0), (top_pad, 0), (0, right_pad)), mode='constant', constant_values=0)
+            right_img = np.lib.pad(right_img, ((0, 0), (top_pad, 0), (0, right_pad)), mode='constant',
+                                   constant_values=0)
+            # pad disparity gt
+            if disparity is not None:
+                assert len(disparity.shape) == 2
+                disparity = np.lib.pad(disparity, ((top_pad, 0), (0, right_pad)), mode='constant', constant_values=0)
+
+            if disparity is not None:
+                return {"left": left_img,
+                        "right": right_img,
+                        "disparity": disparity,
+                        "top_pad": top_pad,
+                        "right_pad": right_pad}
+            else:
+                return {"left": left_img,
+                        "right": right_img,
+                        "top_pad": top_pad,
+                        "right_pad": right_pad}
--- a/datasets/sceneflow_dataset.py
+++ b/datasets/sceneflow_dataset.py
@ -0,0 +1,75 @@
+import os
+import random
+from torch.utils.data import Dataset
+from PIL import Image
+import numpy as np
+from datasets.data_io import get_transform, read_all_lines, pfm_imread
+
+
+class SceneFlowDatset(Dataset):
+    def __init__(self, datapath, list_filename, training):
+        self.datapath = datapath
+        self.left_filenames, self.right_filenames, self.disp_filenames = self.load_path(list_filename)
+        self.training = training
+
+    def load_path(self, list_filename):
+        lines = read_all_lines(list_filename)
+        splits = [line.split() for line in lines]
+        left_images = [x[0] for x in splits]
+        right_images = [x[1] for x in splits]
+        disp_images = [x[2] for x in splits]
+        return left_images, right_images, disp_images
+
+    def load_image(self, filename):
+        return Image.open(filename).convert('RGB')
+
+    def load_disp(self, filename):
+        data, scale = pfm_imread(filename)
+        data = np.ascontiguousarray(data, dtype=np.float32)
+        return data
+
+    def __len__(self):
+        return len(self.left_filenames)
+
+    def __getitem__(self, index):
+        left_img = self.load_image(os.path.join(self.datapath, self.left_filenames[index]))
+        right_img = self.load_image(os.path.join(self.datapath, self.right_filenames[index]))
+        disparity = self.load_disp(os.path.join(self.datapath, self.disp_filenames[index]))
+
+        if self.training:
+            w, h = left_img.size
+            crop_w, crop_h = 512, 256
+
+            x1 = random.randint(0, w - crop_w)
+            y1 = random.randint(0, h - crop_h)
+
+            # random crop
+            left_img = left_img.crop((x1, y1, x1 + crop_w, y1 + crop_h))
+            right_img = right_img.crop((x1, y1, x1 + crop_w, y1 + crop_h))
+            disparity = disparity[y1:y1 + crop_h, x1:x1 + crop_w]
+
+            # to tensor, normalize
+            processed = get_transform()
+            left_img = processed(left_img)
+            right_img = processed(right_img)
+
+            return {"left": left_img,
+                    "right": right_img,
+                    "disparity": disparity}
+        else:
+            w, h = left_img.size
+            crop_w, crop_h = 960, 512
+
+            left_img = left_img.crop((w - crop_w, h - crop_h, w, h))
+            right_img = right_img.crop((w - crop_w, h - crop_h, w, h))
+            disparity = disparity[h - crop_h:h, w - crop_w: w]
+
+            processed = get_transform()
+            left_img = processed(left_img)
+            right_img = processed(right_img)
+
+            return {"left": left_img,
+                    "right": right_img,
+                    "disparity": disparity,
+                    "top_pad": 0,
+                    "right_pad": 0}
--- a/filenames/kitti12_test.txt
+++ b/filenames/kitti12_test.txt
@ -0,0 +1,195 @@
+testing/colored_0/000000_10.png testing/colored_1/000000_10.png
+testing/colored_0/000001_10.png testing/colored_1/000001_10.png
+testing/colored_0/000002_10.png testing/colored_1/000002_10.png
+testing/colored_0/000003_10.png testing/colored_1/000003_10.png
+testing/colored_0/000004_10.png testing/colored_1/000004_10.png
+testing/colored_0/000005_10.png testing/colored_1/000005_10.png
+testing/colored_0/000006_10.png testing/colored_1/000006_10.png
+testing/colored_0/000007_10.png testing/colored_1/000007_10.png
+testing/colored_0/000008_10.png testing/colored_1/000008_10.png
+testing/colored_0/000009_10.png testing/colored_1/000009_10.png
+testing/colored_0/000010_10.png testing/colored_1/000010_10.png
+testing/colored_0/000011_10.png testing/colored_1/000011_10.png
+testing/colored_0/000012_10.png testing/colored_1/000012_10.png
+testing/colored_0/000013_10.png testing/colored_1/000013_10.png
+testing/colored_0/000014_10.png testing/colored_1/000014_10.png
+testing/colored_0/000015_10.png testing/colored_1/000015_10.png
+testing/colored_0/000016_10.png testing/colored_1/000016_10.png
+testing/colored_0/000017_10.png testing/colored_1/000017_10.png
+testing/colored_0/000018_10.png testing/colored_1/000018_10.png
+testing/colored_0/000019_10.png testing/colored_1/000019_10.png
+testing/colored_0/000020_10.png testing/colored_1/000020_10.png
+testing/colored_0/000021_10.png testing/colored_1/000021_10.png
+testing/colored_0/000022_10.png testing/colored_1/000022_10.png
+testing/colored_0/000023_10.png testing/colored_1/000023_10.png
+testing/colored_0/000024_10.png testing/colored_1/000024_10.png
+testing/colored_0/000025_10.png testing/colored_1/000025_10.png
+testing/colored_0/000026_10.png testing/colored_1/000026_10.png
+testing/colored_0/000027_10.png testing/colored_1/000027_10.png
+testing/colored_0/000028_10.png testing/colored_1/000028_10.png
+testing/colored_0/000029_10.png testing/colored_1/000029_10.png
+testing/colored_0/000030_10.png testing/colored_1/000030_10.png
+testing/colored_0/000031_10.png testing/colored_1/000031_10.png
+testing/colored_0/000032_10.png testing/colored_1/000032_10.png
+testing/colored_0/000033_10.png testing/colored_1/000033_10.png
+testing/colored_0/000034_10.png testing/colored_1/000034_10.png
+testing/colored_0/000035_10.png testing/colored_1/000035_10.png
+testing/colored_0/000036_10.png testing/colored_1/000036_10.png
+testing/colored_0/000037_10.png testing/colored_1/000037_10.png
+testing/colored_0/000038_10.png testing/colored_1/000038_10.png
+testing/colored_0/000039_10.png testing/colored_1/000039_10.png
+testing/colored_0/000040_10.png testing/colored_1/000040_10.png
+testing/colored_0/000041_10.png testing/colored_1/000041_10.png
+testing/colored_0/000042_10.png testing/colored_1/000042_10.png
+testing/colored_0/000043_10.png testing/colored_1/000043_10.png
+testing/colored_0/000044_10.png testing/colored_1/000044_10.png
+testing/colored_0/000045_10.png testing/colored_1/000045_10.png
+testing/colored_0/000046_10.png testing/colored_1/000046_10.png
+testing/colored_0/000047_10.png testing/colored_1/000047_10.png
+testing/colored_0/000048_10.png testing/colored_1/000048_10.png
+testing/colored_0/000049_10.png testing/colored_1/000049_10.png
+testing/colored_0/000050_10.png testing/colored_1/000050_10.png
+testing/colored_0/000051_10.png testing/colored_1/000051_10.png
+testing/colored_0/000052_10.png testing/colored_1/000052_10.png
+testing/colored_0/000053_10.png testing/colored_1/000053_10.png
+testing/colored_0/000054_10.png testing/colored_1/000054_10.png
+testing/colored_0/000055_10.png testing/colored_1/000055_10.png
+testing/colored_0/000056_10.png testing/colored_1/000056_10.png
+testing/colored_0/000057_10.png testing/colored_1/000057_10.png
+testing/colored_0/000058_10.png testing/colored_1/000058_10.png
+testing/colored_0/000059_10.png testing/colored_1/000059_10.png
+testing/colored_0/000060_10.png testing/colored_1/000060_10.png
+testing/colored_0/000061_10.png testing/colored_1/000061_10.png
+testing/colored_0/000062_10.png testing/colored_1/000062_10.png
+testing/colored_0/000063_10.png testing/colored_1/000063_10.png
+testing/colored_0/000064_10.png testing/colored_1/000064_10.png
+testing/colored_0/000065_10.png testing/colored_1/000065_10.png
+testing/colored_0/000066_10.png testing/colored_1/000066_10.png
+testing/colored_0/000067_10.png testing/colored_1/000067_10.png
+testing/colored_0/000068_10.png testing/colored_1/000068_10.png
+testing/colored_0/000069_10.png testing/colored_1/000069_10.png
+testing/colored_0/000070_10.png testing/colored_1/000070_10.png
+testing/colored_0/000071_10.png testing/colored_1/000071_10.png
+testing/colored_0/000072_10.png testing/colored_1/000072_10.png
+testing/colored_0/000073_10.png testing/colored_1/000073_10.png
+testing/colored_0/000074_10.png testing/colored_1/000074_10.png
+testing/colored_0/000075_10.png testing/colored_1/000075_10.png
+testing/colored_0/000076_10.png testing/colored_1/000076_10.png
+testing/colored_0/000077_10.png testing/colored_1/000077_10.png
+testing/colored_0/000078_10.png testing/colored_1/000078_10.png
+testing/colored_0/000079_10.png testing/colored_1/000079_10.png
+testing/colored_0/000080_10.png testing/colored_1/000080_10.png
+testing/colored_0/000081_10.png testing/colored_1/000081_10.png
+testing/colored_0/000082_10.png testing/colored_1/000082_10.png
+testing/colored_0/000083_10.png testing/colored_1/000083_10.png
+testing/colored_0/000084_10.png testing/colored_1/000084_10.png
+testing/colored_0/000085_10.png testing/colored_1/000085_10.png
+testing/colored_0/000086_10.png testing/colored_1/000086_10.png
+testing/colored_0/000087_10.png testing/colored_1/000087_10.png
+testing/colored_0/000088_10.png testing/colored_1/000088_10.png
+testing/colored_0/000089_10.png testing/colored_1/000089_10.png
+testing/colored_0/000090_10.png testing/colored_1/000090_10.png
+testing/colored_0/000091_10.png testing/colored_1/000091_10.png
+testing/colored_0/000092_10.png testing/colored_1/000092_10.png
+testing/colored_0/000093_10.png testing/colored_1/000093_10.png
+testing/colored_0/000094_10.png testing/colored_1/000094_10.png
+testing/colored_0/000095_10.png testing/colored_1/000095_10.png
+testing/colored_0/000096_10.png testing/colored_1/000096_10.png
+testing/colored_0/000097_10.png testing/colored_1/000097_10.png
+testing/colored_0/000098_10.png testing/colored_1/000098_10.png
+testing/colored_0/000099_10.png testing/colored_1/000099_10.png
+testing/colored_0/000100_10.png testing/colored_1/000100_10.png
+testing/colored_0/000101_10.png testing/colored_1/000101_10.png
+testing/colored_0/000102_10.png testing/colored_1/000102_10.png
+testing/colored_0/000103_10.png testing/colored_1/000103_10.png
+testing/colored_0/000104_10.png testing/colored_1/000104_10.png
+testing/colored_0/000105_10.png testing/colored_1/000105_10.png
+testing/colored_0/000106_10.png testing/colored_1/000106_10.png
+testing/colored_0/000107_10.png testing/colored_1/000107_10.png
+testing/colored_0/000108_10.png testing/colored_1/000108_10.png
+testing/colored_0/000109_10.png testing/colored_1/000109_10.png
+testing/colored_0/000110_10.png testing/colored_1/000110_10.png
+testing/colored_0/000111_10.png testing/colored_1/000111_10.png
+testing/colored_0/000112_10.png testing/colored_1/000112_10.png
+testing/colored_0/000113_10.png testing/colored_1/000113_10.png
+testing/colored_0/000114_10.png testing/colored_1/000114_10.png
+testing/colored_0/000115_10.png testing/colored_1/000115_10.png
+testing/colored_0/000116_10.png testing/colored_1/000116_10.png
+testing/colored_0/000117_10.png testing/colored_1/000117_10.png
+testing/colored_0/000118_10.png testing/colored_1/000118_10.png
+testing/colored_0/000119_10.png testing/colored_1/000119_10.png
+testing/colored_0/000120_10.png testing/colored_1/000120_10.png
+testing/colored_0/000121_10.png testing/colored_1/000121_10.png
+testing/colored_0/000122_10.png testing/colored_1/000122_10.png
+testing/colored_0/000123_10.png testing/colored_1/000123_10.png
+testing/colored_0/000124_10.png testing/colored_1/000124_10.png
+testing/colored_0/000125_10.png testing/colored_1/000125_10.png
+testing/colored_0/000126_10.png testing/colored_1/000126_10.png
+testing/colored_0/000127_10.png testing/colored_1/000127_10.png
+testing/colored_0/000128_10.png testing/colored_1/000128_10.png
+testing/colored_0/000129_10.png testing/colored_1/000129_10.png
+testing/colored_0/000130_10.png testing/colored_1/000130_10.png
+testing/colored_0/000131_10.png testing/colored_1/000131_10.png
+testing/colored_0/000132_10.png testing/colored_1/000132_10.png
+testing/colored_0/000133_10.png testing/colored_1/000133_10.png
+testing/colored_0/000134_10.png testing/colored_1/000134_10.png
+testing/colored_0/000135_10.png testing/colored_1/000135_10.png
+testing/colored_0/000136_10.png testing/colored_1/000136_10.png
+testing/colored_0/000137_10.png testing/colored_1/000137_10.png
+testing/colored_0/000138_10.png testing/colored_1/000138_10.png
+testing/colored_0/000139_10.png testing/colored_1/000139_10.png
+testing/colored_0/000140_10.png testing/colored_1/000140_10.png
+testing/colored_0/000141_10.png testing/colored_1/000141_10.png
+testing/colored_0/000142_10.png testing/colored_1/000142_10.png
+testing/colored_0/000143_10.png testing/colored_1/000143_10.png
+testing/colored_0/000144_10.png testing/colored_1/000144_10.png
+testing/colored_0/000145_10.png testing/colored_1/000145_10.png
+testing/colored_0/000146_10.png testing/colored_1/000146_10.png
+testing/colored_0/000147_10.png testing/colored_1/000147_10.png
+testing/colored_0/000148_10.png testing/colored_1/000148_10.png
+testing/colored_0/000149_10.png testing/colored_1/000149_10.png
+testing/colored_0/000150_10.png testing/colored_1/000150_10.png
+testing/colored_0/000151_10.png testing/colored_1/000151_10.png
+testing/colored_0/000152_10.png testing/colored_1/000152_10.png
+testing/colored_0/000153_10.png testing/colored_1/000153_10.png
+testing/colored_0/000154_10.png testing/colored_1/000154_10.png
+testing/colored_0/000155_10.png testing/colored_1/000155_10.png
+testing/colored_0/000156_10.png testing/colored_1/000156_10.png
+testing/colored_0/000157_10.png testing/colored_1/000157_10.png
+testing/colored_0/000158_10.png testing/colored_1/000158_10.png
+testing/colored_0/000159_10.png testing/colored_1/000159_10.png
+testing/colored_0/000160_10.png testing/colored_1/000160_10.png
+testing/colored_0/000161_10.png testing/colored_1/000161_10.png
+testing/colored_0/000162_10.png testing/colored_1/000162_10.png
+testing/colored_0/000163_10.png testing/colored_1/000163_10.png
+testing/colored_0/000164_10.png testing/colored_1/000164_10.png
+testing/colored_0/000165_10.png testing/colored_1/000165_10.png
+testing/colored_0/000166_10.png testing/colored_1/000166_10.png
+testing/colored_0/000167_10.png testing/colored_1/000167_10.png
+testing/colored_0/000168_10.png testing/colored_1/000168_10.png
+testing/colored_0/000169_10.png testing/colored_1/000169_10.png
+testing/colored_0/000170_10.png testing/colored_1/000170_10.png
+testing/colored_0/000171_10.png testing/colored_1/000171_10.png
+testing/colored_0/000172_10.png testing/colored_1/000172_10.png
+testing/colored_0/000173_10.png testing/colored_1/000173_10.png
+testing/colored_0/000174_10.png testing/colored_1/000174_10.png
+testing/colored_0/000175_10.png testing/colored_1/000175_10.png
+testing/colored_0/000176_10.png testing/colored_1/000176_10.png
+testing/colored_0/000177_10.png testing/colored_1/000177_10.png
+testing/colored_0/000178_10.png testing/colored_1/000178_10.png
+testing/colored_0/000179_10.png testing/colored_1/000179_10.png
+testing/colored_0/000180_10.png testing/colored_1/000180_10.png
+testing/colored_0/000181_10.png testing/colored_1/000181_10.png
+testing/colored_0/000182_10.png testing/colored_1/000182_10.png
+testing/colored_0/000183_10.png testing/colored_1/000183_10.png
+testing/colored_0/000184_10.png testing/colored_1/000184_10.png
+testing/colored_0/000185_10.png testing/colored_1/000185_10.png
+testing/colored_0/000186_10.png testing/colored_1/000186_10.png
+testing/colored_0/000187_10.png testing/colored_1/000187_10.png
+testing/colored_0/000188_10.png testing/colored_1/000188_10.png
+testing/colored_0/000189_10.png testing/colored_1/000189_10.png
+testing/colored_0/000190_10.png testing/colored_1/000190_10.png
+testing/colored_0/000191_10.png testing/colored_1/000191_10.png
+testing/colored_0/000192_10.png testing/colored_1/000192_10.png
+testing/colored_0/000193_10.png testing/colored_1/000193_10.png
+testing/colored_0/000194_10.png testing/colored_1/000194_10.png
--- a/filenames/kitti12_train.txt
+++ b/filenames/kitti12_train.txt
@ -0,0 +1,180 @@
+training/colored_0/000000_10.png training/colored_1/000000_10.png training/disp_occ/000000_10.png
+training/colored_0/000001_10.png training/colored_1/000001_10.png training/disp_occ/000001_10.png
+training/colored_0/000002_10.png training/colored_1/000002_10.png training/disp_occ/000002_10.png
+training/colored_0/000003_10.png training/colored_1/000003_10.png training/disp_occ/000003_10.png
+training/colored_0/000004_10.png training/colored_1/000004_10.png training/disp_occ/000004_10.png
+training/colored_0/000005_10.png training/colored_1/000005_10.png training/disp_occ/000005_10.png
+training/colored_0/000006_10.png training/colored_1/000006_10.png training/disp_occ/000006_10.png
+training/colored_0/000007_10.png training/colored_1/000007_10.png training/disp_occ/000007_10.png
+training/colored_0/000008_10.png training/colored_1/000008_10.png training/disp_occ/000008_10.png
+training/colored_0/000009_10.png training/colored_1/000009_10.png training/disp_occ/000009_10.png
+training/colored_0/000010_10.png training/colored_1/000010_10.png training/disp_occ/000010_10.png
+training/colored_0/000011_10.png training/colored_1/000011_10.png training/disp_occ/000011_10.png
+training/colored_0/000012_10.png training/colored_1/000012_10.png training/disp_occ/000012_10.png
+training/colored_0/000013_10.png training/colored_1/000013_10.png training/disp_occ/000013_10.png
+training/colored_0/000014_10.png training/colored_1/000014_10.png training/disp_occ/000014_10.png
+training/colored_0/000015_10.png training/colored_1/000015_10.png training/disp_occ/000015_10.png
+training/colored_0/000016_10.png training/colored_1/000016_10.png training/disp_occ/000016_10.png
+training/colored_0/000017_10.png training/colored_1/000017_10.png training/disp_occ/000017_10.png
+training/colored_0/000018_10.png training/colored_1/000018_10.png training/disp_occ/000018_10.png
+training/colored_0/000019_10.png training/colored_1/000019_10.png training/disp_occ/000019_10.png
+training/colored_0/000020_10.png training/colored_1/000020_10.png training/disp_occ/000020_10.png
+training/colored_0/000021_10.png training/colored_1/000021_10.png training/disp_occ/000021_10.png
+training/colored_0/000022_10.png training/colored_1/000022_10.png training/disp_occ/000022_10.png
+training/colored_0/000023_10.png training/colored_1/000023_10.png training/disp_occ/000023_10.png
+training/colored_0/000024_10.png training/colored_1/000024_10.png training/disp_occ/000024_10.png
+training/colored_0/000025_10.png training/colored_1/000025_10.png training/disp_occ/000025_10.png
+training/colored_0/000026_10.png training/colored_1/000026_10.png training/disp_occ/000026_10.png
+training/colored_0/000027_10.png training/colored_1/000027_10.png training/disp_occ/000027_10.png
+training/colored_0/000028_10.png training/colored_1/000028_10.png training/disp_occ/000028_10.png
+training/colored_0/000029_10.png training/colored_1/000029_10.png training/disp_occ/000029_10.png
+training/colored_0/000030_10.png training/colored_1/000030_10.png training/disp_occ/000030_10.png
+training/colored_0/000031_10.png training/colored_1/000031_10.png training/disp_occ/000031_10.png
+training/colored_0/000032_10.png training/colored_1/000032_10.png training/disp_occ/000032_10.png
+training/colored_0/000033_10.png training/colored_1/000033_10.png training/disp_occ/000033_10.png
+training/colored_0/000034_10.png training/colored_1/000034_10.png training/disp_occ/000034_10.png
+training/colored_0/000035_10.png training/colored_1/000035_10.png training/disp_occ/000035_10.png
+training/colored_0/000036_10.png training/colored_1/000036_10.png training/disp_occ/000036_10.png
+training/colored_0/000037_10.png training/colored_1/000037_10.png training/disp_occ/000037_10.png
+training/colored_0/000038_10.png training/colored_1/000038_10.png training/disp_occ/000038_10.png
+training/colored_0/000039_10.png training/colored_1/000039_10.png training/disp_occ/000039_10.png
+training/colored_0/000040_10.png training/colored_1/000040_10.png training/disp_occ/000040_10.png
+training/colored_0/000041_10.png training/colored_1/000041_10.png training/disp_occ/000041_10.png
+training/colored_0/000042_10.png training/colored_1/000042_10.png training/disp_occ/000042_10.png
+training/colored_0/000043_10.png training/colored_1/000043_10.png training/disp_occ/000043_10.png
+training/colored_0/000044_10.png training/colored_1/000044_10.png training/disp_occ/000044_10.png
+training/colored_0/000045_10.png training/colored_1/000045_10.png training/disp_occ/000045_10.png
+training/colored_0/000046_10.png training/colored_1/000046_10.png training/disp_occ/000046_10.png
+training/colored_0/000047_10.png training/colored_1/000047_10.png training/disp_occ/000047_10.png
+training/colored_0/000048_10.png training/colored_1/000048_10.png training/disp_occ/000048_10.png
+training/colored_0/000049_10.png training/colored_1/000049_10.png training/disp_occ/000049_10.png
+training/colored_0/000050_10.png training/colored_1/000050_10.png training/disp_occ/000050_10.png
+training/colored_0/000051_10.png training/colored_1/000051_10.png training/disp_occ/000051_10.png
+training/colored_0/000052_10.png training/colored_1/000052_10.png training/disp_occ/000052_10.png
+training/colored_0/000053_10.png training/colored_1/000053_10.png training/disp_occ/000053_10.png
+training/colored_0/000054_10.png training/colored_1/000054_10.png training/disp_occ/000054_10.png
+training/colored_0/000055_10.png training/colored_1/000055_10.png training/disp_occ/000055_10.png
+training/colored_0/000056_10.png training/colored_1/000056_10.png training/disp_occ/000056_10.png
+training/colored_0/000057_10.png training/colored_1/000057_10.png training/disp_occ/000057_10.png
+training/colored_0/000058_10.png training/colored_1/000058_10.png training/disp_occ/000058_10.png
+training/colored_0/000059_10.png training/colored_1/000059_10.png training/disp_occ/000059_10.png
+training/colored_0/000060_10.png training/colored_1/000060_10.png training/disp_occ/000060_10.png
+training/colored_0/000061_10.png training/colored_1/000061_10.png training/disp_occ/000061_10.png
+training/colored_0/000062_10.png training/colored_1/000062_10.png training/disp_occ/000062_10.png
+training/colored_0/000063_10.png training/colored_1/000063_10.png training/disp_occ/000063_10.png
+training/colored_0/000064_10.png training/colored_1/000064_10.png training/disp_occ/000064_10.png
+training/colored_0/000065_10.png training/colored_1/000065_10.png training/disp_occ/000065_10.png
+training/colored_0/000066_10.png training/colored_1/000066_10.png training/disp_occ/000066_10.png
+training/colored_0/000067_10.png training/colored_1/000067_10.png training/disp_occ/000067_10.png
+training/colored_0/000068_10.png training/colored_1/000068_10.png training/disp_occ/000068_10.png
+training/colored_0/000069_10.png training/colored_1/000069_10.png training/disp_occ/000069_10.png
+training/colored_0/000070_10.png training/colored_1/000070_10.png training/disp_occ/000070_10.png
+training/colored_0/000071_10.png training/colored_1/000071_10.png training/disp_occ/000071_10.png
+training/colored_0/000072_10.png training/colored_1/000072_10.png training/disp_occ/000072_10.png
+training/colored_0/000073_10.png training/colored_1/000073_10.png training/disp_occ/000073_10.png
+training/colored_0/000074_10.png training/colored_1/000074_10.png training/disp_occ/000074_10.png
+training/colored_0/000075_10.png training/colored_1/000075_10.png training/disp_occ/000075_10.png
+training/colored_0/000076_10.png training/colored_1/000076_10.png training/disp_occ/000076_10.png
+training/colored_0/000077_10.png training/colored_1/000077_10.png training/disp_occ/000077_10.png
+training/colored_0/000078_10.png training/colored_1/000078_10.png training/disp_occ/000078_10.png
+training/colored_0/000079_10.png training/colored_1/000079_10.png training/disp_occ/000079_10.png
+training/colored_0/000080_10.png training/colored_1/000080_10.png training/disp_occ/000080_10.png
+training/colored_0/000081_10.png training/colored_1/000081_10.png training/disp_occ/000081_10.png
+training/colored_0/000082_10.png training/colored_1/000082_10.png training/disp_occ/000082_10.png
+training/colored_0/000083_10.png training/colored_1/000083_10.png training/disp_occ/000083_10.png
+training/colored_0/000084_10.png training/colored_1/000084_10.png training/disp_occ/000084_10.png
+training/colored_0/000085_10.png training/colored_1/000085_10.png training/disp_occ/000085_10.png
+training/colored_0/000086_10.png training/colored_1/000086_10.png training/disp_occ/000086_10.png
+training/colored_0/000087_10.png training/colored_1/000087_10.png training/disp_occ/000087_10.png
+training/colored_0/000088_10.png training/colored_1/000088_10.png training/disp_occ/000088_10.png
+training/colored_0/000089_10.png training/colored_1/000089_10.png training/disp_occ/000089_10.png
+training/colored_0/000090_10.png training/colored_1/000090_10.png training/disp_occ/000090_10.png
+training/colored_0/000091_10.png training/colored_1/000091_10.png training/disp_occ/000091_10.png
+training/colored_0/000092_10.png training/colored_1/000092_10.png training/disp_occ/000092_10.png
+training/colored_0/000093_10.png training/colored_1/000093_10.png training/disp_occ/000093_10.png
+training/colored_0/000094_10.png training/colored_1/000094_10.png training/disp_occ/000094_10.png
+training/colored_0/000095_10.png training/colored_1/000095_10.png training/disp_occ/000095_10.png
+training/colored_0/000096_10.png training/colored_1/000096_10.png training/disp_occ/000096_10.png
+training/colored_0/000097_10.png training/colored_1/000097_10.png training/disp_occ/000097_10.png
+training/colored_0/000098_10.png training/colored_1/000098_10.png training/disp_occ/000098_10.png
+training/colored_0/000099_10.png training/colored_1/000099_10.png training/disp_occ/000099_10.png
+training/colored_0/000100_10.png training/colored_1/000100_10.png training/disp_occ/000100_10.png
+training/colored_0/000101_10.png training/colored_1/000101_10.png training/disp_occ/000101_10.png
+training/colored_0/000102_10.png training/colored_1/000102_10.png training/disp_occ/000102_10.png
+training/colored_0/000103_10.png training/colored_1/000103_10.png training/disp_occ/000103_10.png
+training/colored_0/000104_10.png training/colored_1/000104_10.png training/disp_occ/000104_10.png
+training/colored_0/000105_10.png training/colored_1/000105_10.png training/disp_occ/000105_10.png
+training/colored_0/000106_10.png training/colored_1/000106_10.png training/disp_occ/000106_10.png
+training/colored_0/000107_10.png training/colored_1/000107_10.png training/disp_occ/000107_10.png
+training/colored_0/000108_10.png training/colored_1/000108_10.png training/disp_occ/000108_10.png
+training/colored_0/000109_10.png training/colored_1/000109_10.png training/disp_occ/000109_10.png
+training/colored_0/000110_10.png training/colored_1/000110_10.png training/disp_occ/000110_10.png
+training/colored_0/000111_10.png training/colored_1/000111_10.png training/disp_occ/000111_10.png
+training/colored_0/000112_10.png training/colored_1/000112_10.png training/disp_occ/000112_10.png
+training/colored_0/000113_10.png training/colored_1/000113_10.png training/disp_occ/000113_10.png
+training/colored_0/000114_10.png training/colored_1/000114_10.png training/disp_occ/000114_10.png
+training/colored_0/000115_10.png training/colored_1/000115_10.png training/disp_occ/000115_10.png
+training/colored_0/000116_10.png training/colored_1/000116_10.png training/disp_occ/000116_10.png
+training/colored_0/000117_10.png training/colored_1/000117_10.png training/disp_occ/000117_10.png
+training/colored_0/000118_10.png training/colored_1/000118_10.png training/disp_occ/000118_10.png
+training/colored_0/000119_10.png training/colored_1/000119_10.png training/disp_occ/000119_10.png
+training/colored_0/000120_10.png training/colored_1/000120_10.png training/disp_occ/000120_10.png
+training/colored_0/000121_10.png training/colored_1/000121_10.png training/disp_occ/000121_10.png
+training/colored_0/000122_10.png training/colored_1/000122_10.png training/disp_occ/000122_10.png
+training/colored_0/000123_10.png training/colored_1/000123_10.png training/disp_occ/000123_10.png
+training/colored_0/000124_10.png training/colored_1/000124_10.png training/disp_occ/000124_10.png
+training/colored_0/000125_10.png training/colored_1/000125_10.png training/disp_occ/000125_10.png
+training/colored_0/000126_10.png training/colored_1/000126_10.png training/disp_occ/000126_10.png
+training/colored_0/000127_10.png training/colored_1/000127_10.png training/disp_occ/000127_10.png
+training/colored_0/000128_10.png training/colored_1/000128_10.png training/disp_occ/000128_10.png
+training/colored_0/000129_10.png training/colored_1/000129_10.png training/disp_occ/000129_10.png
+training/colored_0/000130_10.png training/colored_1/000130_10.png training/disp_occ/000130_10.png
+training/colored_0/000131_10.png training/colored_1/000131_10.png training/disp_occ/000131_10.png
+training/colored_0/000132_10.png training/colored_1/000132_10.png training/disp_occ/000132_10.png
+training/colored_0/000133_10.png training/colored_1/000133_10.png training/disp_occ/000133_10.png
+training/colored_0/000134_10.png training/colored_1/000134_10.png training/disp_occ/000134_10.png
+training/colored_0/000135_10.png training/colored_1/000135_10.png training/disp_occ/000135_10.png
+training/colored_0/000136_10.png training/colored_1/000136_10.png training/disp_occ/000136_10.png
+training/colored_0/000137_10.png training/colored_1/000137_10.png training/disp_occ/000137_10.png
+training/colored_0/000138_10.png training/colored_1/000138_10.png training/disp_occ/000138_10.png
+training/colored_0/000139_10.png training/colored_1/000139_10.png training/disp_occ/000139_10.png
+training/colored_0/000140_10.png training/colored_1/000140_10.png training/disp_occ/000140_10.png
+training/colored_0/000141_10.png training/colored_1/000141_10.png training/disp_occ/000141_10.png
+training/colored_0/000142_10.png training/colored_1/000142_10.png training/disp_occ/000142_10.png
+training/colored_0/000143_10.png training/colored_1/000143_10.png training/disp_occ/000143_10.png
+training/colored_0/000144_10.png training/colored_1/000144_10.png training/disp_occ/000144_10.png
+training/colored_0/000145_10.png training/colored_1/000145_10.png training/disp_occ/000145_10.png
+training/colored_0/000146_10.png training/colored_1/000146_10.png training/disp_occ/000146_10.png
+training/colored_0/000147_10.png training/colored_1/000147_10.png training/disp_occ/000147_10.png
+training/colored_0/000148_10.png training/colored_1/000148_10.png training/disp_occ/000148_10.png
+training/colored_0/000149_10.png training/colored_1/000149_10.png training/disp_occ/000149_10.png
+training/colored_0/000150_10.png training/colored_1/000150_10.png training/disp_occ/000150_10.png
+training/colored_0/000151_10.png training/colored_1/000151_10.png training/disp_occ/000151_10.png
+training/colored_0/000152_10.png training/colored_1/000152_10.png training/disp_occ/000152_10.png
+training/colored_0/000153_10.png training/colored_1/000153_10.png training/disp_occ/000153_10.png
+training/colored_0/000154_10.png training/colored_1/000154_10.png training/disp_occ/000154_10.png
+training/colored_0/000155_10.png training/colored_1/000155_10.png training/disp_occ/000155_10.png
+training/colored_0/000156_10.png training/colored_1/000156_10.png training/disp_occ/000156_10.png
+training/colored_0/000157_10.png training/colored_1/000157_10.png training/disp_occ/000157_10.png
+training/colored_0/000158_10.png training/colored_1/000158_10.png training/disp_occ/000158_10.png
+training/colored_0/000159_10.png training/colored_1/000159_10.png training/disp_occ/000159_10.png
+training/colored_0/000160_10.png training/colored_1/000160_10.png training/disp_occ/000160_10.png
+training/colored_0/000161_10.png training/colored_1/000161_10.png training/disp_occ/000161_10.png
+training/colored_0/000162_10.png training/colored_1/000162_10.png training/disp_occ/000162_10.png
+training/colored_0/000163_10.png training/colored_1/000163_10.png training/disp_occ/000163_10.png
+training/colored_0/000164_10.png training/colored_1/000164_10.png training/disp_occ/000164_10.png
+training/colored_0/000165_10.png training/colored_1/000165_10.png training/disp_occ/000165_10.png
+training/colored_0/000166_10.png training/colored_1/000166_10.png training/disp_occ/000166_10.png
+training/colored_0/000167_10.png training/colored_1/000167_10.png training/disp_occ/000167_10.png
+training/colored_0/000168_10.png training/colored_1/000168_10.png training/disp_occ/000168_10.png
+training/colored_0/000169_10.png training/colored_1/000169_10.png training/disp_occ/000169_10.png
+training/colored_0/000170_10.png training/colored_1/000170_10.png training/disp_occ/000170_10.png
+training/colored_0/000171_10.png training/colored_1/000171_10.png training/disp_occ/000171_10.png
+training/colored_0/000172_10.png training/colored_1/000172_10.png training/disp_occ/000172_10.png
+training/colored_0/000173_10.png training/colored_1/000173_10.png training/disp_occ/000173_10.png
+training/colored_0/000174_10.png training/colored_1/000174_10.png training/disp_occ/000174_10.png
+training/colored_0/000175_10.png training/colored_1/000175_10.png training/disp_occ/000175_10.png
+training/colored_0/000176_10.png training/colored_1/000176_10.png training/disp_occ/000176_10.png
+training/colored_0/000177_10.png training/colored_1/000177_10.png training/disp_occ/000177_10.png
+training/colored_0/000178_10.png training/colored_1/000178_10.png training/disp_occ/000178_10.png
+training/colored_0/000179_10.png training/colored_1/000179_10.png training/disp_occ/000179_10.png
--- a/filenames/kitti12_val.txt
+++ b/filenames/kitti12_val.txt
@ -0,0 +1,14 @@
+training/colored_0/000180_10.png training/colored_1/000180_10.png training/disp_occ/000180_10.png
+training/colored_0/000181_10.png training/colored_1/000181_10.png training/disp_occ/000181_10.png
+training/colored_0/000182_10.png training/colored_1/000182_10.png training/disp_occ/000182_10.png
+training/colored_0/000183_10.png training/colored_1/000183_10.png training/disp_occ/000183_10.png
+training/colored_0/000184_10.png training/colored_1/000184_10.png training/disp_occ/000184_10.png
+training/colored_0/000185_10.png training/colored_1/000185_10.png training/disp_occ/000185_10.png
+training/colored_0/000186_10.png training/colored_1/000186_10.png training/disp_occ/000186_10.png
+training/colored_0/000187_10.png training/colored_1/000187_10.png training/disp_occ/000187_10.png
+training/colored_0/000188_10.png training/colored_1/000188_10.png training/disp_occ/000188_10.png
+training/colored_0/000189_10.png training/colored_1/000189_10.png training/disp_occ/000189_10.png
+training/colored_0/000190_10.png training/colored_1/000190_10.png training/disp_occ/000190_10.png
+training/colored_0/000191_10.png training/colored_1/000191_10.png training/disp_occ/000191_10.png
+training/colored_0/000192_10.png training/colored_1/000192_10.png training/disp_occ/000192_10.png
+training/colored_0/000193_10.png training/colored_1/000193_10.png training/disp_occ/000193_10.png
--- a/filenames/kitti15_test.txt
+++ b/filenames/kitti15_test.txt
@ -0,0 +1,200 @@
+testing/image_2/000000_10.png testing/image_3/000000_10.png
+testing/image_2/000001_10.png testing/image_3/000001_10.png
+testing/image_2/000002_10.png testing/image_3/000002_10.png
+testing/image_2/000003_10.png testing/image_3/000003_10.png
+testing/image_2/000004_10.png testing/image_3/000004_10.png
+testing/image_2/000005_10.png testing/image_3/000005_10.png
+testing/image_2/000006_10.png testing/image_3/000006_10.png
+testing/image_2/000007_10.png testing/image_3/000007_10.png
+testing/image_2/000008_10.png testing/image_3/000008_10.png
+testing/image_2/000009_10.png testing/image_3/000009_10.png
+testing/image_2/000010_10.png testing/image_3/000010_10.png
+testing/image_2/000011_10.png testing/image_3/000011_10.png
+testing/image_2/000012_10.png testing/image_3/000012_10.png
+testing/image_2/000013_10.png testing/image_3/000013_10.png
+testing/image_2/000014_10.png testing/image_3/000014_10.png
+testing/image_2/000015_10.png testing/image_3/000015_10.png
+testing/image_2/000016_10.png testing/image_3/000016_10.png
+testing/image_2/000017_10.png testing/image_3/000017_10.png
+testing/image_2/000018_10.png testing/image_3/000018_10.png
+testing/image_2/000019_10.png testing/image_3/000019_10.png
+testing/image_2/000020_10.png testing/image_3/000020_10.png
+testing/image_2/000021_10.png testing/image_3/000021_10.png
+testing/image_2/000022_10.png testing/image_3/000022_10.png
+testing/image_2/000023_10.png testing/image_3/000023_10.png
+testing/image_2/000024_10.png testing/image_3/000024_10.png
+testing/image_2/000025_10.png testing/image_3/000025_10.png
+testing/image_2/000026_10.png testing/image_3/000026_10.png
+testing/image_2/000027_10.png testing/image_3/000027_10.png
+testing/image_2/000028_10.png testing/image_3/000028_10.png
+testing/image_2/000029_10.png testing/image_3/000029_10.png
+testing/image_2/000030_10.png testing/image_3/000030_10.png
+testing/image_2/000031_10.png testing/image_3/000031_10.png
+testing/image_2/000032_10.png testing/image_3/000032_10.png
+testing/image_2/000033_10.png testing/image_3/000033_10.png
+testing/image_2/000034_10.png testing/image_3/000034_10.png
+testing/image_2/000035_10.png testing/image_3/000035_10.png
+testing/image_2/000036_10.png testing/image_3/000036_10.png
+testing/image_2/000037_10.png testing/image_3/000037_10.png
+testing/image_2/000038_10.png testing/image_3/000038_10.png
+testing/image_2/000039_10.png testing/image_3/000039_10.png
+testing/image_2/000040_10.png testing/image_3/000040_10.png
+testing/image_2/000041_10.png testing/image_3/000041_10.png
+testing/image_2/000042_10.png testing/image_3/000042_10.png
+testing/image_2/000043_10.png testing/image_3/000043_10.png
+testing/image_2/000044_10.png testing/image_3/000044_10.png
+testing/image_2/000045_10.png testing/image_3/000045_10.png
+testing/image_2/000046_10.png testing/image_3/000046_10.png
+testing/image_2/000047_10.png testing/image_3/000047_10.png
+testing/image_2/000048_10.png testing/image_3/000048_10.png
+testing/image_2/000049_10.png testing/image_3/000049_10.png
+testing/image_2/000050_10.png testing/image_3/000050_10.png
+testing/image_2/000051_10.png testing/image_3/000051_10.png
+testing/image_2/000052_10.png testing/image_3/000052_10.png
+testing/image_2/000053_10.png testing/image_3/000053_10.png
+testing/image_2/000054_10.png testing/image_3/000054_10.png
+testing/image_2/000055_10.png testing/image_3/000055_10.png
+testing/image_2/000056_10.png testing/image_3/000056_10.png
+testing/image_2/000057_10.png testing/image_3/000057_10.png
+testing/image_2/000058_10.png testing/image_3/000058_10.png
+testing/image_2/000059_10.png testing/image_3/000059_10.png
+testing/image_2/000060_10.png testing/image_3/000060_10.png
+testing/image_2/000061_10.png testing/image_3/000061_10.png
+testing/image_2/000062_10.png testing/image_3/000062_10.png
+testing/image_2/000063_10.png testing/image_3/000063_10.png
+testing/image_2/000064_10.png testing/image_3/000064_10.png
+testing/image_2/000065_10.png testing/image_3/000065_10.png
+testing/image_2/000066_10.png testing/image_3/000066_10.png
+testing/image_2/000067_10.png testing/image_3/000067_10.png
+testing/image_2/000068_10.png testing/image_3/000068_10.png
+testing/image_2/000069_10.png testing/image_3/000069_10.png
+testing/image_2/000070_10.png testing/image_3/000070_10.png
+testing/image_2/000071_10.png testing/image_3/000071_10.png
+testing/image_2/000072_10.png testing/image_3/000072_10.png
+testing/image_2/000073_10.png testing/image_3/000073_10.png
+testing/image_2/000074_10.png testing/image_3/000074_10.png
+testing/image_2/000075_10.png testing/image_3/000075_10.png
+testing/image_2/000076_10.png testing/image_3/000076_10.png
+testing/image_2/000077_10.png testing/image_3/000077_10.png
+testing/image_2/000078_10.png testing/image_3/000078_10.png
+testing/image_2/000079_10.png testing/image_3/000079_10.png
+testing/image_2/000080_10.png testing/image_3/000080_10.png
+testing/image_2/000081_10.png testing/image_3/000081_10.png
+testing/image_2/000082_10.png testing/image_3/000082_10.png
+testing/image_2/000083_10.png testing/image_3/000083_10.png
+testing/image_2/000084_10.png testing/image_3/000084_10.png
+testing/image_2/000085_10.png testing/image_3/000085_10.png
+testing/image_2/000086_10.png testing/image_3/000086_10.png
+testing/image_2/000087_10.png testing/image_3/000087_10.png
+testing/image_2/000088_10.png testing/image_3/000088_10.png
+testing/image_2/000089_10.png testing/image_3/000089_10.png
+testing/image_2/000090_10.png testing/image_3/000090_10.png
+testing/image_2/000091_10.png testing/image_3/000091_10.png
+testing/image_2/000092_10.png testing/image_3/000092_10.png
+testing/image_2/000093_10.png testing/image_3/000093_10.png
+testing/image_2/000094_10.png testing/image_3/000094_10.png
+testing/image_2/000095_10.png testing/image_3/000095_10.png
+testing/image_2/000096_10.png testing/image_3/000096_10.png
+testing/image_2/000097_10.png testing/image_3/000097_10.png
+testing/image_2/000098_10.png testing/image_3/000098_10.png
+testing/image_2/000099_10.png testing/image_3/000099_10.png
+testing/image_2/000100_10.png testing/image_3/000100_10.png
+testing/image_2/000101_10.png testing/image_3/000101_10.png
+testing/image_2/000102_10.png testing/image_3/000102_10.png
+testing/image_2/000103_10.png testing/image_3/000103_10.png
+testing/image_2/000104_10.png testing/image_3/000104_10.png
+testing/image_2/000105_10.png testing/image_3/000105_10.png
+testing/image_2/000106_10.png testing/image_3/000106_10.png
+testing/image_2/000107_10.png testing/image_3/000107_10.png
+testing/image_2/000108_10.png testing/image_3/000108_10.png
+testing/image_2/000109_10.png testing/image_3/000109_10.png
+testing/image_2/000110_10.png testing/image_3/000110_10.png
+testing/image_2/000111_10.png testing/image_3/000111_10.png
+testing/image_2/000112_10.png testing/image_3/000112_10.png
+testing/image_2/000113_10.png testing/image_3/000113_10.png
+testing/image_2/000114_10.png testing/image_3/000114_10.png
+testing/image_2/000115_10.png testing/image_3/000115_10.png
+testing/image_2/000116_10.png testing/image_3/000116_10.png
+testing/image_2/000117_10.png testing/image_3/000117_10.png
+testing/image_2/000118_10.png testing/image_3/000118_10.png
+testing/image_2/000119_10.png testing/image_3/000119_10.png
+testing/image_2/000120_10.png testing/image_3/000120_10.png
+testing/image_2/000121_10.png testing/image_3/000121_10.png
+testing/image_2/000122_10.png testing/image_3/000122_10.png
+testing/image_2/000123_10.png testing/image_3/000123_10.png
+testing/image_2/000124_10.png testing/image_3/000124_10.png
+testing/image_2/000125_10.png testing/image_3/000125_10.png
+testing/image_2/000126_10.png testing/image_3/000126_10.png
+testing/image_2/000127_10.png testing/image_3/000127_10.png
+testing/image_2/000128_10.png testing/image_3/000128_10.png
+testing/image_2/000129_10.png testing/image_3/000129_10.png
+testing/image_2/000130_10.png testing/image_3/000130_10.png
+testing/image_2/000131_10.png testing/image_3/000131_10.png
+testing/image_2/000132_10.png testing/image_3/000132_10.png
+testing/image_2/000133_10.png testing/image_3/000133_10.png
+testing/image_2/000134_10.png testing/image_3/000134_10.png
+testing/image_2/000135_10.png testing/image_3/000135_10.png
+testing/image_2/000136_10.png testing/image_3/000136_10.png
+testing/image_2/000137_10.png testing/image_3/000137_10.png
+testing/image_2/000138_10.png testing/image_3/000138_10.png
+testing/image_2/000139_10.png testing/image_3/000139_10.png
+testing/image_2/000140_10.png testing/image_3/000140_10.png
+testing/image_2/000141_10.png testing/image_3/000141_10.png
+testing/image_2/000142_10.png testing/image_3/000142_10.png
+testing/image_2/000143_10.png testing/image_3/000143_10.png
+testing/image_2/000144_10.png testing/image_3/000144_10.png
+testing/image_2/000145_10.png testing/image_3/000145_10.png
+testing/image_2/000146_10.png testing/image_3/000146_10.png
+testing/image_2/000147_10.png testing/image_3/000147_10.png
+testing/image_2/000148_10.png testing/image_3/000148_10.png
+testing/image_2/000149_10.png testing/image_3/000149_10.png
+testing/image_2/000150_10.png testing/image_3/000150_10.png
+testing/image_2/000151_10.png testing/image_3/000151_10.png
+testing/image_2/000152_10.png testing/image_3/000152_10.png
+testing/image_2/000153_10.png testing/image_3/000153_10.png
+testing/image_2/000154_10.png testing/image_3/000154_10.png
+testing/image_2/000155_10.png testing/image_3/000155_10.png
+testing/image_2/000156_10.png testing/image_3/000156_10.png
+testing/image_2/000157_10.png testing/image_3/000157_10.png
+testing/image_2/000158_10.png testing/image_3/000158_10.png
+testing/image_2/000159_10.png testing/image_3/000159_10.png
+testing/image_2/000160_10.png testing/image_3/000160_10.png
+testing/image_2/000161_10.png testing/image_3/000161_10.png
+testing/image_2/000162_10.png testing/image_3/000162_10.png
+testing/image_2/000163_10.png testing/image_3/000163_10.png
+testing/image_2/000164_10.png testing/image_3/000164_10.png
+testing/image_2/000165_10.png testing/image_3/000165_10.png
+testing/image_2/000166_10.png testing/image_3/000166_10.png
+testing/image_2/000167_10.png testing/image_3/000167_10.png
+testing/image_2/000168_10.png testing/image_3/000168_10.png
+testing/image_2/000169_10.png testing/image_3/000169_10.png
+testing/image_2/000170_10.png testing/image_3/000170_10.png
+testing/image_2/000171_10.png testing/image_3/000171_10.png
+testing/image_2/000172_10.png testing/image_3/000172_10.png
+testing/image_2/000173_10.png testing/image_3/000173_10.png
+testing/image_2/000174_10.png testing/image_3/000174_10.png
+testing/image_2/000175_10.png testing/image_3/000175_10.png
+testing/image_2/000176_10.png testing/image_3/000176_10.png
+testing/image_2/000177_10.png testing/image_3/000177_10.png
+testing/image_2/000178_10.png testing/image_3/000178_10.png
+testing/image_2/000179_10.png testing/image_3/000179_10.png
+testing/image_2/000180_10.png testing/image_3/000180_10.png
+testing/image_2/000181_10.png testing/image_3/000181_10.png
+testing/image_2/000182_10.png testing/image_3/000182_10.png
+testing/image_2/000183_10.png testing/image_3/000183_10.png
+testing/image_2/000184_10.png testing/image_3/000184_10.png
+testing/image_2/000185_10.png testing/image_3/000185_10.png
+testing/image_2/000186_10.png testing/image_3/000186_10.png
+testing/image_2/000187_10.png testing/image_3/000187_10.png
+testing/image_2/000188_10.png testing/image_3/000188_10.png
+testing/image_2/000189_10.png testing/image_3/000189_10.png
+testing/image_2/000190_10.png testing/image_3/000190_10.png
+testing/image_2/000191_10.png testing/image_3/000191_10.png
+testing/image_2/000192_10.png testing/image_3/000192_10.png
+testing/image_2/000193_10.png testing/image_3/000193_10.png
+testing/image_2/000194_10.png testing/image_3/000194_10.png
+testing/image_2/000195_10.png testing/image_3/000195_10.png
+testing/image_2/000196_10.png testing/image_3/000196_10.png
+testing/image_2/000197_10.png testing/image_3/000197_10.png
+testing/image_2/000198_10.png testing/image_3/000198_10.png
+testing/image_2/000199_10.png testing/image_3/000199_10.png
--- a/filenames/kitti15_train.txt
+++ b/filenames/kitti15_train.txt
@ -0,0 +1,180 @@
+training/image_2/000000_10.png training/image_3/000000_10.png training/disp_occ_0/000000_10.png
+training/image_2/000002_10.png training/image_3/000002_10.png training/disp_occ_0/000002_10.png
+training/image_2/000003_10.png training/image_3/000003_10.png training/disp_occ_0/000003_10.png
+training/image_2/000004_10.png training/image_3/000004_10.png training/disp_occ_0/000004_10.png
+training/image_2/000005_10.png training/image_3/000005_10.png training/disp_occ_0/000005_10.png
+training/image_2/000007_10.png training/image_3/000007_10.png training/disp_occ_0/000007_10.png
+training/image_2/000008_10.png training/image_3/000008_10.png training/disp_occ_0/000008_10.png
+training/image_2/000009_10.png training/image_3/000009_10.png training/disp_occ_0/000009_10.png
+training/image_2/000010_10.png training/image_3/000010_10.png training/disp_occ_0/000010_10.png
+training/image_2/000011_10.png training/image_3/000011_10.png training/disp_occ_0/000011_10.png
+training/image_2/000012_10.png training/image_3/000012_10.png training/disp_occ_0/000012_10.png
+training/image_2/000013_10.png training/image_3/000013_10.png training/disp_occ_0/000013_10.png
+training/image_2/000014_10.png training/image_3/000014_10.png training/disp_occ_0/000014_10.png
+training/image_2/000015_10.png training/image_3/000015_10.png training/disp_occ_0/000015_10.png
+training/image_2/000016_10.png training/image_3/000016_10.png training/disp_occ_0/000016_10.png
+training/image_2/000017_10.png training/image_3/000017_10.png training/disp_occ_0/000017_10.png
+training/image_2/000018_10.png training/image_3/000018_10.png training/disp_occ_0/000018_10.png
+training/image_2/000019_10.png training/image_3/000019_10.png training/disp_occ_0/000019_10.png
+training/image_2/000020_10.png training/image_3/000020_10.png training/disp_occ_0/000020_10.png
+training/image_2/000021_10.png training/image_3/000021_10.png training/disp_occ_0/000021_10.png
+training/image_2/000022_10.png training/image_3/000022_10.png training/disp_occ_0/000022_10.png
+training/image_2/000023_10.png training/image_3/000023_10.png training/disp_occ_0/000023_10.png
+training/image_2/000024_10.png training/image_3/000024_10.png training/disp_occ_0/000024_10.png
+training/image_2/000025_10.png training/image_3/000025_10.png training/disp_occ_0/000025_10.png
+training/image_2/000027_10.png training/image_3/000027_10.png training/disp_occ_0/000027_10.png
+training/image_2/000028_10.png training/image_3/000028_10.png training/disp_occ_0/000028_10.png
+training/image_2/000029_10.png training/image_3/000029_10.png training/disp_occ_0/000029_10.png
+training/image_2/000030_10.png training/image_3/000030_10.png training/disp_occ_0/000030_10.png
+training/image_2/000031_10.png training/image_3/000031_10.png training/disp_occ_0/000031_10.png
+training/image_2/000032_10.png training/image_3/000032_10.png training/disp_occ_0/000032_10.png
+training/image_2/000033_10.png training/image_3/000033_10.png training/disp_occ_0/000033_10.png
+training/image_2/000034_10.png training/image_3/000034_10.png training/disp_occ_0/000034_10.png
+training/image_2/000035_10.png training/image_3/000035_10.png training/disp_occ_0/000035_10.png
+training/image_2/000036_10.png training/image_3/000036_10.png training/disp_occ_0/000036_10.png
+training/image_2/000037_10.png training/image_3/000037_10.png training/disp_occ_0/000037_10.png
+training/image_2/000039_10.png training/image_3/000039_10.png training/disp_occ_0/000039_10.png
+training/image_2/000040_10.png training/image_3/000040_10.png training/disp_occ_0/000040_10.png
+training/image_2/000041_10.png training/image_3/000041_10.png training/disp_occ_0/000041_10.png
+training/image_2/000042_10.png training/image_3/000042_10.png training/disp_occ_0/000042_10.png
+training/image_2/000044_10.png training/image_3/000044_10.png training/disp_occ_0/000044_10.png
+training/image_2/000045_10.png training/image_3/000045_10.png training/disp_occ_0/000045_10.png
+training/image_2/000046_10.png training/image_3/000046_10.png training/disp_occ_0/000046_10.png
+training/image_2/000047_10.png training/image_3/000047_10.png training/disp_occ_0/000047_10.png
+training/image_2/000048_10.png training/image_3/000048_10.png training/disp_occ_0/000048_10.png
+training/image_2/000050_10.png training/image_3/000050_10.png training/disp_occ_0/000050_10.png
+training/image_2/000051_10.png training/image_3/000051_10.png training/disp_occ_0/000051_10.png
+training/image_2/000052_10.png training/image_3/000052_10.png training/disp_occ_0/000052_10.png
+training/image_2/000053_10.png training/image_3/000053_10.png training/disp_occ_0/000053_10.png
+training/image_2/000054_10.png training/image_3/000054_10.png training/disp_occ_0/000054_10.png
+training/image_2/000055_10.png training/image_3/000055_10.png training/disp_occ_0/000055_10.png
+training/image_2/000056_10.png training/image_3/000056_10.png training/disp_occ_0/000056_10.png
+training/image_2/000057_10.png training/image_3/000057_10.png training/disp_occ_0/000057_10.png
+training/image_2/000058_10.png training/image_3/000058_10.png training/disp_occ_0/000058_10.png
+training/image_2/000059_10.png training/image_3/000059_10.png training/disp_occ_0/000059_10.png
+training/image_2/000060_10.png training/image_3/000060_10.png training/disp_occ_0/000060_10.png
+training/image_2/000061_10.png training/image_3/000061_10.png training/disp_occ_0/000061_10.png
+training/image_2/000062_10.png training/image_3/000062_10.png training/disp_occ_0/000062_10.png
+training/image_2/000063_10.png training/image_3/000063_10.png training/disp_occ_0/000063_10.png
+training/image_2/000064_10.png training/image_3/000064_10.png training/disp_occ_0/000064_10.png
+training/image_2/000065_10.png training/image_3/000065_10.png training/disp_occ_0/000065_10.png
+training/image_2/000066_10.png training/image_3/000066_10.png training/disp_occ_0/000066_10.png
+training/image_2/000068_10.png training/image_3/000068_10.png training/disp_occ_0/000068_10.png
+training/image_2/000069_10.png training/image_3/000069_10.png training/disp_occ_0/000069_10.png
+training/image_2/000070_10.png training/image_3/000070_10.png training/disp_occ_0/000070_10.png
+training/image_2/000071_10.png training/image_3/000071_10.png training/disp_occ_0/000071_10.png
+training/image_2/000072_10.png training/image_3/000072_10.png training/disp_occ_0/000072_10.png
+training/image_2/000073_10.png training/image_3/000073_10.png training/disp_occ_0/000073_10.png
+training/image_2/000074_10.png training/image_3/000074_10.png training/disp_occ_0/000074_10.png
+training/image_2/000075_10.png training/image_3/000075_10.png training/disp_occ_0/000075_10.png
+training/image_2/000076_10.png training/image_3/000076_10.png training/disp_occ_0/000076_10.png
+training/image_2/000077_10.png training/image_3/000077_10.png training/disp_occ_0/000077_10.png
+training/image_2/000078_10.png training/image_3/000078_10.png training/disp_occ_0/000078_10.png
+training/image_2/000079_10.png training/image_3/000079_10.png training/disp_occ_0/000079_10.png
+training/image_2/000080_10.png training/image_3/000080_10.png training/disp_occ_0/000080_10.png
+training/image_2/000082_10.png training/image_3/000082_10.png training/disp_occ_0/000082_10.png
+training/image_2/000083_10.png training/image_3/000083_10.png training/disp_occ_0/000083_10.png
+training/image_2/000084_10.png training/image_3/000084_10.png training/disp_occ_0/000084_10.png
+training/image_2/000085_10.png training/image_3/000085_10.png training/disp_occ_0/000085_10.png
+training/image_2/000086_10.png training/image_3/000086_10.png training/disp_occ_0/000086_10.png
+training/image_2/000087_10.png training/image_3/000087_10.png training/disp_occ_0/000087_10.png
+training/image_2/000088_10.png training/image_3/000088_10.png training/disp_occ_0/000088_10.png
+training/image_2/000090_10.png training/image_3/000090_10.png training/disp_occ_0/000090_10.png
+training/image_2/000091_10.png training/image_3/000091_10.png training/disp_occ_0/000091_10.png
+training/image_2/000092_10.png training/image_3/000092_10.png training/disp_occ_0/000092_10.png
+training/image_2/000093_10.png training/image_3/000093_10.png training/disp_occ_0/000093_10.png
+training/image_2/000094_10.png training/image_3/000094_10.png training/disp_occ_0/000094_10.png
+training/image_2/000095_10.png training/image_3/000095_10.png training/disp_occ_0/000095_10.png
+training/image_2/000096_10.png training/image_3/000096_10.png training/disp_occ_0/000096_10.png
+training/image_2/000097_10.png training/image_3/000097_10.png training/disp_occ_0/000097_10.png
+training/image_2/000098_10.png training/image_3/000098_10.png training/disp_occ_0/000098_10.png
+training/image_2/000099_10.png training/image_3/000099_10.png training/disp_occ_0/000099_10.png
+training/image_2/000100_10.png training/image_3/000100_10.png training/disp_occ_0/000100_10.png
+training/image_2/000101_10.png training/image_3/000101_10.png training/disp_occ_0/000101_10.png
+training/image_2/000102_10.png training/image_3/000102_10.png training/disp_occ_0/000102_10.png
+training/image_2/000103_10.png training/image_3/000103_10.png training/disp_occ_0/000103_10.png
+training/image_2/000104_10.png training/image_3/000104_10.png training/disp_occ_0/000104_10.png
+training/image_2/000105_10.png training/image_3/000105_10.png training/disp_occ_0/000105_10.png
+training/image_2/000106_10.png training/image_3/000106_10.png training/disp_occ_0/000106_10.png
+training/image_2/000107_10.png training/image_3/000107_10.png training/disp_occ_0/000107_10.png
+training/image_2/000108_10.png training/image_3/000108_10.png training/disp_occ_0/000108_10.png
+training/image_2/000110_10.png training/image_3/000110_10.png training/disp_occ_0/000110_10.png
+training/image_2/000111_10.png training/image_3/000111_10.png training/disp_occ_0/000111_10.png
+training/image_2/000112_10.png training/image_3/000112_10.png training/disp_occ_0/000112_10.png
+training/image_2/000113_10.png training/image_3/000113_10.png training/disp_occ_0/000113_10.png
+training/image_2/000114_10.png training/image_3/000114_10.png training/disp_occ_0/000114_10.png
+training/image_2/000115_10.png training/image_3/000115_10.png training/disp_occ_0/000115_10.png
+training/image_2/000116_10.png training/image_3/000116_10.png training/disp_occ_0/000116_10.png
+training/image_2/000117_10.png training/image_3/000117_10.png training/disp_occ_0/000117_10.png
+training/image_2/000118_10.png training/image_3/000118_10.png training/disp_occ_0/000118_10.png
+training/image_2/000119_10.png training/image_3/000119_10.png training/disp_occ_0/000119_10.png
+training/image_2/000120_10.png training/image_3/000120_10.png training/disp_occ_0/000120_10.png
+training/image_2/000121_10.png training/image_3/000121_10.png training/disp_occ_0/000121_10.png
+training/image_2/000123_10.png training/image_3/000123_10.png training/disp_occ_0/000123_10.png
+training/image_2/000124_10.png training/image_3/000124_10.png training/disp_occ_0/000124_10.png
+training/image_2/000125_10.png training/image_3/000125_10.png training/disp_occ_0/000125_10.png
+training/image_2/000126_10.png training/image_3/000126_10.png training/disp_occ_0/000126_10.png
+training/image_2/000127_10.png training/image_3/000127_10.png training/disp_occ_0/000127_10.png
+training/image_2/000128_10.png training/image_3/000128_10.png training/disp_occ_0/000128_10.png
+training/image_2/000130_10.png training/image_3/000130_10.png training/disp_occ_0/000130_10.png
+training/image_2/000131_10.png training/image_3/000131_10.png training/disp_occ_0/000131_10.png
+training/image_2/000133_10.png training/image_3/000133_10.png training/disp_occ_0/000133_10.png
+training/image_2/000134_10.png training/image_3/000134_10.png training/disp_occ_0/000134_10.png
+training/image_2/000135_10.png training/image_3/000135_10.png training/disp_occ_0/000135_10.png
+training/image_2/000136_10.png training/image_3/000136_10.png training/disp_occ_0/000136_10.png
+training/image_2/000137_10.png training/image_3/000137_10.png training/disp_occ_0/000137_10.png
+training/image_2/000138_10.png training/image_3/000138_10.png training/disp_occ_0/000138_10.png
+training/image_2/000139_10.png training/image_3/000139_10.png training/disp_occ_0/000139_10.png
+training/image_2/000140_10.png training/image_3/000140_10.png training/disp_occ_0/000140_10.png
+training/image_2/000142_10.png training/image_3/000142_10.png training/disp_occ_0/000142_10.png
+training/image_2/000143_10.png training/image_3/000143_10.png training/disp_occ_0/000143_10.png
+training/image_2/000144_10.png training/image_3/000144_10.png training/disp_occ_0/000144_10.png
+training/image_2/000145_10.png training/image_3/000145_10.png training/disp_occ_0/000145_10.png
+training/image_2/000146_10.png training/image_3/000146_10.png training/disp_occ_0/000146_10.png
+training/image_2/000147_10.png training/image_3/000147_10.png training/disp_occ_0/000147_10.png
+training/image_2/000148_10.png training/image_3/000148_10.png training/disp_occ_0/000148_10.png
+training/image_2/000149_10.png training/image_3/000149_10.png training/disp_occ_0/000149_10.png
+training/image_2/000150_10.png training/image_3/000150_10.png training/disp_occ_0/000150_10.png
+training/image_2/000151_10.png training/image_3/000151_10.png training/disp_occ_0/000151_10.png
+training/image_2/000153_10.png training/image_3/000153_10.png training/disp_occ_0/000153_10.png
+training/image_2/000154_10.png training/image_3/000154_10.png training/disp_occ_0/000154_10.png
+training/image_2/000155_10.png training/image_3/000155_10.png training/disp_occ_0/000155_10.png
+training/image_2/000156_10.png training/image_3/000156_10.png training/disp_occ_0/000156_10.png
+training/image_2/000157_10.png training/image_3/000157_10.png training/disp_occ_0/000157_10.png
+training/image_2/000158_10.png training/image_3/000158_10.png training/disp_occ_0/000158_10.png
+training/image_2/000160_10.png training/image_3/000160_10.png training/disp_occ_0/000160_10.png
+training/image_2/000161_10.png training/image_3/000161_10.png training/disp_occ_0/000161_10.png
+training/image_2/000162_10.png training/image_3/000162_10.png training/disp_occ_0/000162_10.png
+training/image_2/000163_10.png training/image_3/000163_10.png training/disp_occ_0/000163_10.png
+training/image_2/000164_10.png training/image_3/000164_10.png training/disp_occ_0/000164_10.png
+training/image_2/000165_10.png training/image_3/000165_10.png training/disp_occ_0/000165_10.png
+training/image_2/000166_10.png training/image_3/000166_10.png training/disp_occ_0/000166_10.png
+training/image_2/000167_10.png training/image_3/000167_10.png training/disp_occ_0/000167_10.png
+training/image_2/000168_10.png training/image_3/000168_10.png training/disp_occ_0/000168_10.png
+training/image_2/000169_10.png training/image_3/000169_10.png training/disp_occ_0/000169_10.png
+training/image_2/000170_10.png training/image_3/000170_10.png training/disp_occ_0/000170_10.png
+training/image_2/000172_10.png training/image_3/000172_10.png training/disp_occ_0/000172_10.png
+training/image_2/000173_10.png training/image_3/000173_10.png training/disp_occ_0/000173_10.png
+training/image_2/000174_10.png training/image_3/000174_10.png training/disp_occ_0/000174_10.png
+training/image_2/000175_10.png training/image_3/000175_10.png training/disp_occ_0/000175_10.png
+training/image_2/000176_10.png training/image_3/000176_10.png training/disp_occ_0/000176_10.png
+training/image_2/000177_10.png training/image_3/000177_10.png training/disp_occ_0/000177_10.png
+training/image_2/000178_10.png training/image_3/000178_10.png training/disp_occ_0/000178_10.png
+training/image_2/000180_10.png training/image_3/000180_10.png training/disp_occ_0/000180_10.png
+training/image_2/000181_10.png training/image_3/000181_10.png training/disp_occ_0/000181_10.png
+training/image_2/000182_10.png training/image_3/000182_10.png training/disp_occ_0/000182_10.png
+training/image_2/000183_10.png training/image_3/000183_10.png training/disp_occ_0/000183_10.png
+training/image_2/000185_10.png training/image_3/000185_10.png training/disp_occ_0/000185_10.png
+training/image_2/000186_10.png training/image_3/000186_10.png training/disp_occ_0/000186_10.png
+training/image_2/000188_10.png training/image_3/000188_10.png training/disp_occ_0/000188_10.png
+training/image_2/000189_10.png training/image_3/000189_10.png training/disp_occ_0/000189_10.png
+training/image_2/000190_10.png training/image_3/000190_10.png training/disp_occ_0/000190_10.png
+training/image_2/000191_10.png training/image_3/000191_10.png training/disp_occ_0/000191_10.png
+training/image_2/000192_10.png training/image_3/000192_10.png training/disp_occ_0/000192_10.png
+training/image_2/000193_10.png training/image_3/000193_10.png training/disp_occ_0/000193_10.png
+training/image_2/000194_10.png training/image_3/000194_10.png training/disp_occ_0/000194_10.png
+training/image_2/000195_10.png training/image_3/000195_10.png training/disp_occ_0/000195_10.png
+training/image_2/000196_10.png training/image_3/000196_10.png training/disp_occ_0/000196_10.png
+training/image_2/000197_10.png training/image_3/000197_10.png training/disp_occ_0/000197_10.png
+training/image_2/000198_10.png training/image_3/000198_10.png training/disp_occ_0/000198_10.png
+training/image_2/000199_10.png training/image_3/000199_10.png training/disp_occ_0/000199_10.png
--- a/filenames/kitti15_val.txt
+++ b/filenames/kitti15_val.txt
@ -0,0 +1,20 @@
+training/image_2/000001_10.png training/image_3/000001_10.png training/disp_occ_0/000001_10.png
+training/image_2/000006_10.png training/image_3/000006_10.png training/disp_occ_0/000006_10.png
+training/image_2/000026_10.png training/image_3/000026_10.png training/disp_occ_0/000026_10.png
+training/image_2/000038_10.png training/image_3/000038_10.png training/disp_occ_0/000038_10.png
+training/image_2/000043_10.png training/image_3/000043_10.png training/disp_occ_0/000043_10.png
+training/image_2/000049_10.png training/image_3/000049_10.png training/disp_occ_0/000049_10.png
+training/image_2/000067_10.png training/image_3/000067_10.png training/disp_occ_0/000067_10.png
+training/image_2/000081_10.png training/image_3/000081_10.png training/disp_occ_0/000081_10.png
+training/image_2/000089_10.png training/image_3/000089_10.png training/disp_occ_0/000089_10.png
+training/image_2/000109_10.png training/image_3/000109_10.png training/disp_occ_0/000109_10.png
+training/image_2/000122_10.png training/image_3/000122_10.png training/disp_occ_0/000122_10.png
+training/image_2/000129_10.png training/image_3/000129_10.png training/disp_occ_0/000129_10.png
+training/image_2/000132_10.png training/image_3/000132_10.png training/disp_occ_0/000132_10.png
+training/image_2/000141_10.png training/image_3/000141_10.png training/disp_occ_0/000141_10.png
+training/image_2/000152_10.png training/image_3/000152_10.png training/disp_occ_0/000152_10.png
+training/image_2/000159_10.png training/image_3/000159_10.png training/disp_occ_0/000159_10.png
+training/image_2/000171_10.png training/image_3/000171_10.png training/disp_occ_0/000171_10.png
+training/image_2/000179_10.png training/image_3/000179_10.png training/disp_occ_0/000179_10.png
+training/image_2/000184_10.png training/image_3/000184_10.png training/disp_occ_0/000184_10.png
+training/image_2/000187_10.png training/image_3/000187_10.png training/disp_occ_0/000187_10.png
--- a/filenames/sceneflow_test.txt
+++ b/filenames/sceneflow_test.txt
--- a/filenames/sceneflow_train.txt
+++ b/filenames/sceneflow_train.txt
--- a/kitti12.sh
+++ b/kitti12.sh
@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -x
+DATAPATH="/home/xyguo/data/kitti_2012/"
+python main.py --dataset kitti \
+    --datapath $DATAPATH --trainlist ./filenames/kitti12_train.txt --testlist ./filenames/kitti12_val.txt \
+    --epochs 300 --lrepochs "200:10" \
+    --model gwcnet-gc --logdir ./checkpoints/kitti12/gwcnet-gc --loadckpt ./checkpoints/sceneflow/gwcnet-gc/pretrained.ckpt \
+    --test_batch_size 1
--- a/kitti15.sh
+++ b/kitti15.sh
@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -x
+DATAPATH="/home/xyguo/data/kitti_2015/"
+python main.py --dataset kitti \
+    --datapath $DATAPATH --trainlist ./filenames/kitti15_train.txt --testlist ./filenames/kitti15_val.txt \
+    --epochs 300 --lrepochs "200:10" \
+    --model gwcnet-g --logdir ./checkpoints/kitti15/gwcnet-g --loadckpt ./checkpoints/sceneflow/gwcnet-g/pretrained.ckpt \
+    --test_batch_size 1
--- a/main.py
+++ b/main.py
@ -0,0 +1,202 @@
+from __future__ import print_function, division
+import argparse
+import os
+import torch
+import torch.nn as nn
+import torch.nn.parallel
+import torch.backends.cudnn as cudnn
+import torch.optim as optim
+import torch.utils.data
+from torch.autograd import Variable
+import torchvision.utils as vutils
+import torch.nn.functional as F
+import numpy as np
+import time
+from tensorboardX import SummaryWriter
+from datasets import __datasets__
+from models import __models__
+from models import *
+from utils import *
+from torch.utils.data import DataLoader
+import skimage
+import gc
+import datetime
+import cv2
+
+cudnn.benchmark = True
+
+parser = argparse.ArgumentParser(description='Group-wise Correlation Stereo Network (GwcNet)')
+parser.add_argument('--model', default='gwcnet-g', help='select a model structure', choices=__models__.keys())
+
+parser.add_argument('--dataset', required=True, help='dataset name', choices=__datasets__.keys())
+parser.add_argument('--datapath', required=True, help='data path')
+parser.add_argument('--trainlist', required=True, help='training list')
+parser.add_argument('--testlist', required=True, help='testing list')
+
+parser.add_argument('--lr', type=float, default=0.001, help='base learning rate')
+parser.add_argument('--batch_size', type=int, default=16, help='training batch size')
+parser.add_argument('--test_batch_size', type=int, default=8, help='testing batch size')
+parser.add_argument('--maxdisp', type=int, default=192, help='maximum disparity')
+parser.add_argument('--epochs', type=int, required=True, help='number of epochs to train')
+parser.add_argument('--lrepochs', type=str, required=True, help='the epochs to decay lr: the downscale rate')
+
+parser.add_argument('--logdir', required=True, help='the directory to save logs and checkpoints')
+parser.add_argument('--loadckpt', help='load the weights from a specific checkpoint')
+parser.add_argument('--resume', action='store_true', help='continue training the model')
+parser.add_argument('--seed', type=int, default=1, metavar='S', help='random seed (default: 1)')
+
+parser.add_argument('--summary_freq', type=int, default=20, help='the frequency of saving summary')
+parser.add_argument('--save_freq', type=int, default=1, help='the frequency of saving checkpoint')
+
+# parse arguments, set seeds
+args = parser.parse_args()
+torch.manual_seed(args.seed)
+torch.cuda.manual_seed(args.seed)
+os.makedirs(args.logdir, exist_ok=True)
+
+# create summary logger
+print("creating new summary file")
+logger = SummaryWriter(args.logdir)
+
+# dataset, dataloader
+StereoDataset = __datasets__[args.dataset]
+train_dataset = StereoDataset(args.datapath, args.trainlist, True)
+test_dataset = StereoDataset(args.datapath, args.testlist, False)
+TrainImgLoader = DataLoader(train_dataset, args.batch_size, shuffle=True, num_workers=8, drop_last=True)
+TestImgLoader = DataLoader(test_dataset, args.test_batch_size, shuffle=False, num_workers=4, drop_last=False)
+
+# model, optimizer
+model = __models__[args.model](args.maxdisp)
+model = nn.DataParallel(model)
+model.cuda()
+optimizer = optim.Adam(model.parameters(), lr=args.lr, betas=(0.9, 0.999))
+
+# load parameters
+start_epoch = 0
+if args.resume:
+    # find all checkpoints file and sort according to epoch id
+    all_saved_ckpts = [fn for fn in os.listdir(args.logdir) if fn.endswith(".ckpt")]
+    all_saved_ckpts = sorted(all_saved_ckpts, key=lambda x: int(x.split('_')[-1].split('.')[0]))
+    # use the latest checkpoint file
+    loadckpt = os.path.join(args.logdir, all_saved_ckpts[-1])
+    print("loading the lastest model in logdir: {}".format(loadckpt))
+    state_dict = torch.load(loadckpt)
+    model.load_state_dict(state_dict['model'])
+    optimizer.load_state_dict(state_dict['optimizer'])
+    start_epoch = state_dict['epoch'] + 1
+elif args.loadckpt:
+    # load the checkpoint file specified by args.loadckpt
+    print("loading model {}".format(args.loadckpt))
+    state_dict = torch.load(args.loadckpt)
+    model.load_state_dict(state_dict['model'])
+print("start at epoch {}".format(start_epoch))
+
+
+def train():
+    for epoch_idx in range(start_epoch, args.epochs):
+        adjust_learning_rate(optimizer, epoch_idx, args.lr, args.lrepochs)
+
+        # training
+        for batch_idx, sample in enumerate(TrainImgLoader):
+            global_step = len(TrainImgLoader) * epoch_idx + batch_idx
+            start_time = time.time()
+            do_summary = global_step % args.summary_freq == 0
+            loss, scalar_outputs, image_outputs = train_sample(sample, compute_metrics=do_summary)
+            if do_summary:
+                save_scalars(logger, 'train', scalar_outputs, global_step)
+                save_images(logger, 'train', image_outputs, global_step)
+            del scalar_outputs, image_outputs
+            print('Epoch {}/{}, Iter {}/{}, train loss = {:.3f}, time = {:.3f}'.format(epoch_idx, args.epochs,
+                                                                                       batch_idx,
+                                                                                       len(TrainImgLoader), loss,
+                                                                                       time.time() - start_time))
+        # saving checkpoints
+        if (epoch_idx + 1) % args.save_freq == 0:
+            checkpoint_data = {'epoch': epoch_idx, 'model': model.state_dict(), 'optimizer': optimizer.state_dict()}
+            torch.save(checkpoint_data, "{}/checkpoint_{:0>6}.ckpt".format(args.logdir, epoch_idx))
+        gc.collect()
+
+        # testing
+        avg_test_scalars = AverageMeterDict()
+        for batch_idx, sample in enumerate(TestImgLoader):
+            global_step = len(TrainImgLoader) * epoch_idx + batch_idx
+            start_time = time.time()
+            do_summary = global_step % args.summary_freq == 0
+            loss, scalar_outputs, image_outputs = test_sample(sample, compute_metrics=do_summary)
+            if do_summary:
+                save_scalars(logger, 'test', scalar_outputs, global_step)
+                save_images(logger, 'test', image_outputs, global_step)
+            avg_test_scalars.update(scalar_outputs)
+            del scalar_outputs, image_outputs
+            print('Epoch {}/{}, Iter {}/{}, test loss = {:.3f}, time = {:3f}'.format(epoch_idx, args.epochs,
+                                                                                     batch_idx,
+                                                                                     len(TestImgLoader), loss,
+                                                                                     time.time() - start_time))
+        avg_test_scalars = avg_test_scalars.mean()
+        save_scalars(logger, 'fulltest', avg_test_scalars, len(TrainImgLoader) * (epoch_idx + 1))
+        print("avg_test_scalars", avg_test_scalars)
+        gc.collect()
+
+
+# train one sample
+def train_sample(sample, compute_metrics=False):
+    model.train()
+
+    imgL, imgR, disp_gt = sample['left'], sample['right'], sample['disparity']
+    imgL = imgL.cuda()
+    imgR = imgR.cuda()
+    disp_gt = disp_gt.cuda()
+
+    optimizer.zero_grad()
+
+    disp_ests = model(imgL, imgR)
+    mask = (disp_gt < args.maxdisp) & (disp_gt > 0)
+    loss = model_loss(disp_ests, disp_gt, mask)
+
+    scalar_outputs = {"loss": loss}
+    image_outputs = {"disp_est": disp_ests, "disp_gt": disp_gt, "imgL": imgL, "imgR": imgR}
+    if compute_metrics:
+        with torch.no_grad():
+            image_outputs["errormap"] = [disp_error_image_func()(disp_est, disp_gt) for disp_est in disp_ests]
+            scalar_outputs["EPE"] = [EPE_metric(disp_est, disp_gt, mask) for disp_est in disp_ests]
+            scalar_outputs["D1"] = [D1_metric(disp_est, disp_gt, mask) for disp_est in disp_ests]
+            scalar_outputs["Thres1"] = [Thres_metric(disp_est, disp_gt, mask, 1.0) for disp_est in disp_ests]
+            scalar_outputs["Thres2"] = [Thres_metric(disp_est, disp_gt, mask, 2.0) for disp_est in disp_ests]
+            scalar_outputs["Thres3"] = [Thres_metric(disp_est, disp_gt, mask, 3.0) for disp_est in disp_ests]
+    loss.backward()
+    optimizer.step()
+
+    return tensor2float(loss), tensor2float(scalar_outputs), image_outputs
+
+
+# test one sample
+@make_nograd_func
+def test_sample(sample, compute_metrics=True):
+    model.eval()
+
+    imgL, imgR, disp_gt = sample['left'], sample['right'], sample['disparity']
+    imgL = imgL.cuda()
+    imgR = imgR.cuda()
+    disp_gt = disp_gt.cuda()
+
+    disp_ests = model(imgL, imgR)
+    mask = (disp_gt < args.maxdisp) & (disp_gt > 0)
+    loss = model_loss(disp_ests, disp_gt, mask)
+
+    scalar_outputs = {"loss": loss}
+    image_outputs = {"disp_est": disp_ests, "disp_gt": disp_gt, "imgL": imgL, "imgR": imgR}
+
+    scalar_outputs["D1"] = [D1_metric(disp_est, disp_gt, mask) for disp_est in disp_ests]
+    scalar_outputs["EPE"] = [EPE_metric(disp_est, disp_gt, mask) for disp_est in disp_ests]
+    scalar_outputs["Thres1"] = [Thres_metric(disp_est, disp_gt, mask, 1.0) for disp_est in disp_ests]
+    scalar_outputs["Thres2"] = [Thres_metric(disp_est, disp_gt, mask, 2.0) for disp_est in disp_ests]
+    scalar_outputs["Thres3"] = [Thres_metric(disp_est, disp_gt, mask, 3.0) for disp_est in disp_ests]
+
+    if compute_metrics:
+        image_outputs["errormap"] = [disp_error_image_func()(disp_est, disp_gt) for disp_est in disp_ests]
+
+    return tensor2float(loss), tensor2float(scalar_outputs), image_outputs
+
+
+if __name__ == '__main__':
+    train()
--- a/models/init.py
+++ b/models/init.py
@ -0,0 +1,7 @@
+from models.gwcnet import GwcNet_G, GwcNet_GC
+from models.loss import model_loss
+
+__models__ = {
+    "gwcnet-g": GwcNet_G,
+    "gwcnet-gc": GwcNet_GC
+}
--- a/models/gwcnet.py
+++ b/models/gwcnet.py
@ -0,0 +1,231 @@
+from __future__ import print_function
+import torch
+import torch.nn as nn
+import torch.utils.data
+from torch.autograd import Variable
+import torch.nn.functional as F
+import math
+from models.submodule import *
+
+
+class feature_extraction(nn.Module):
+    def __init__(self, concat_feature=False, concat_feature_channel=12):
+        super(feature_extraction, self).__init__()
+        self.concat_feature = concat_feature
+
+        self.inplanes = 32
+        self.firstconv = nn.Sequential(convbn(3, 32, 3, 2, 1, 1),
+                                       nn.ReLU(inplace=True),
+                                       convbn(32, 32, 3, 1, 1, 1),
+                                       nn.ReLU(inplace=True),
+                                       convbn(32, 32, 3, 1, 1, 1),
+                                       nn.ReLU(inplace=True))
+
+        self.layer1 = self._make_layer(BasicBlock, 32, 3, 1, 1, 1)
+        self.layer2 = self._make_layer(BasicBlock, 64, 16, 2, 1, 1)
+        self.layer3 = self._make_layer(BasicBlock, 128, 3, 1, 1, 1)
+        self.layer4 = self._make_layer(BasicBlock, 128, 3, 1, 1, 2)
+
+        if self.concat_feature:
+            self.lastconv = nn.Sequential(convbn(320, 128, 3, 1, 1, 1),
+                                          nn.ReLU(inplace=True),
+                                          nn.Conv2d(128, concat_feature_channel, kernel_size=1, padding=0, stride=1,
+                                                    bias=False))
+
+    def _make_layer(self, block, planes, blocks, stride, pad, dilation):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2d(self.inplanes, planes * block.expansion,
+                          kernel_size=1, stride=stride, bias=False),
+                nn.BatchNorm2d(planes * block.expansion), )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample, pad, dilation))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes, 1, None, pad, dilation))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.firstconv(x)
+        x = self.layer1(x)
+        l2 = self.layer2(x)
+        l3 = self.layer3(l2)
+        l4 = self.layer4(l3)
+
+        gwc_feature = torch.cat((l2, l3, l4), dim=1)
+
+        if not self.concat_feature:
+            return {"gwc_feature": gwc_feature}
+        else:
+            concat_feature = self.lastconv(gwc_feature)
+            return {"gwc_feature": gwc_feature, "concat_feature": concat_feature}
+
+
+class hourglass(nn.Module):
+    def __init__(self, in_channels):
+        super(hourglass, self).__init__()
+
+        self.conv1 = nn.Sequential(convbn_3d(in_channels, in_channels * 2, 3, 2, 1),
+                                   nn.ReLU(inplace=True))
+
+        self.conv2 = nn.Sequential(convbn_3d(in_channels * 2, in_channels * 2, 3, 2, 1),
+                                   nn.ReLU(inplace=True))
+
+        self.conv3 = nn.Sequential(convbn_3d(in_channels * 2, in_channels * 4, 3, 2, 1),
+                                   nn.ReLU(inplace=True))
+
+        self.conv4 = nn.Sequential(convbn_3d(in_channels * 4, in_channels * 4, 3, 2, 1),
+                                   nn.ReLU(inplace=True))
+
+        self.conv5 = nn.Sequential(
+            nn.ConvTranspose3d(in_channels * 4, in_channels * 2, 3, padding=1, output_padding=1, stride=2, bias=False),
+            nn.BatchNorm3d(in_channels * 2))
+
+        self.conv6 = nn.Sequential(
+            nn.ConvTranspose3d(in_channels * 2, in_channels, 3, padding=1, output_padding=1, stride=2, bias=False),
+            nn.BatchNorm3d(in_channels))
+
+        self.redir1 = convbn_3d(in_channels, in_channels, kernel_size=1, stride=1, pad=0)
+        self.redir2 = convbn_3d(in_channels * 2, in_channels * 2, kernel_size=1, stride=1, pad=0)
+
+    def forward(self, x):
+        conv1 = self.conv1(x)
+        conv2 = self.conv2(conv1)
+
+        conv3 = self.conv3(conv2)
+        conv4 = self.conv4(conv3)
+
+        conv5 = F.relu(self.conv5(conv4) + self.redir2(conv2), inplace=True)
+        conv6 = F.relu(self.conv6(conv5) + self.redir1(x), inplace=True)
+
+        return conv6
+
+
+class GwcNet(nn.Module):
+    def __init__(self, maxdisp, use_concat_volume=False):
+        super(GwcNet, self).__init__()
+        self.maxdisp = maxdisp
+        self.use_concat_volume = use_concat_volume
+
+        self.num_groups = 40
+
+        if self.use_concat_volume:
+            self.concat_channels = 12
+            self.feature_extraction = feature_extraction(concat_feature=True,
+                                                         concat_feature_channel=self.concat_channels)
+        else:
+            self.concat_channels = 0
+            self.feature_extraction = feature_extraction(concat_feature=False)
+
+        self.dres0 = nn.Sequential(convbn_3d(self.num_groups + self.concat_channels * 2, 32, 3, 1, 1),
+                                   nn.ReLU(inplace=True),
+                                   convbn_3d(32, 32, 3, 1, 1),
+                                   nn.ReLU(inplace=True))
+
+        self.dres1 = nn.Sequential(convbn_3d(32, 32, 3, 1, 1),
+                                   nn.ReLU(inplace=True),
+                                   convbn_3d(32, 32, 3, 1, 1))
+
+        self.dres2 = hourglass(32)
+
+        self.dres3 = hourglass(32)
+
+        self.dres4 = hourglass(32)
+
+        self.classif0 = nn.Sequential(convbn_3d(32, 32, 3, 1, 1),
+                                      nn.ReLU(inplace=True),
+                                      nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
+
+        self.classif1 = nn.Sequential(convbn_3d(32, 32, 3, 1, 1),
+                                      nn.ReLU(inplace=True),
+                                      nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
+
+        self.classif2 = nn.Sequential(convbn_3d(32, 32, 3, 1, 1),
+                                      nn.ReLU(inplace=True),
+                                      nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
+
+        self.classif3 = nn.Sequential(convbn_3d(32, 32, 3, 1, 1),
+                                      nn.ReLU(inplace=True),
+                                      nn.Conv3d(32, 1, kernel_size=3, padding=1, stride=1, bias=False))
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
+                m.weight.data.normal_(0, math.sqrt(2. / n))
+            elif isinstance(m, nn.Conv3d):
+                n = m.kernel_size[0] * m.kernel_size[1] * m.kernel_size[2] * m.out_channels
+                m.weight.data.normal_(0, math.sqrt(2. / n))
+            elif isinstance(m, nn.BatchNorm2d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.BatchNorm3d):
+                m.weight.data.fill_(1)
+                m.bias.data.zero_()
+            elif isinstance(m, nn.Linear):
+                m.bias.data.zero_()
+
+    def forward(self, left, right):
+        features_left = self.feature_extraction(left)
+        features_right = self.feature_extraction(right)
+
+        gwc_volume = build_gwc_volume(features_left["gwc_feature"], features_right["gwc_feature"], self.maxdisp // 4,
+                                      self.num_groups)
+        if self.use_concat_volume:
+            concat_volume = build_concat_volume(features_left["concat_feature"], features_right["concat_feature"],
+                                                self.maxdisp // 4)
+            volume = torch.cat((gwc_volume, concat_volume), 1)
+        else:
+            volume = gwc_volume
+
+        cost0 = self.dres0(volume)
+        cost0 = self.dres1(cost0) + cost0
+
+        out1 = self.dres2(cost0)
+        out2 = self.dres3(out1)
+        out3 = self.dres4(out2)
+
+        if self.training:
+            cost0 = self.classif0(cost0)
+            cost1 = self.classif1(out1)
+            cost2 = self.classif2(out2)
+            cost3 = self.classif3(out3)
+
+            cost0 = F.upsample(cost0, [self.maxdisp, left.size()[2], left.size()[3]], mode='trilinear')
+            cost0 = torch.squeeze(cost0, 1)
+            pred0 = F.softmax(cost0, dim=1)
+            pred0 = disparity_regression(pred0, self.maxdisp)
+
+            cost1 = F.upsample(cost1, [self.maxdisp, left.size()[2], left.size()[3]], mode='trilinear')
+            cost1 = torch.squeeze(cost1, 1)
+            pred1 = F.softmax(cost1, dim=1)
+            pred1 = disparity_regression(pred1, self.maxdisp)
+
+            cost2 = F.upsample(cost2, [self.maxdisp, left.size()[2], left.size()[3]], mode='trilinear')
+            cost2 = torch.squeeze(cost2, 1)
+            pred2 = F.softmax(cost2, dim=1)
+            pred2 = disparity_regression(pred2, self.maxdisp)
+
+            cost3 = F.upsample(cost3, [self.maxdisp, left.size()[2], left.size()[3]], mode='trilinear')
+            cost3 = torch.squeeze(cost3, 1)
+            pred3 = F.softmax(cost3, dim=1)
+            pred3 = disparity_regression(pred3, self.maxdisp)
+            return [pred0, pred1, pred2, pred3]
+
+        else:
+            cost3 = self.classif3(out3)
+            cost3 = F.upsample(cost3, [self.maxdisp, left.size()[2], left.size()[3]], mode='trilinear')
+            cost3 = torch.squeeze(cost3, 1)
+            pred3 = F.softmax(cost3, dim=1)
+            pred3 = disparity_regression(pred3, self.maxdisp)
+            return [pred3]
+
+
+def GwcNet_G(d):
+    return GwcNet(d, use_concat_volume=False)
+
+
+def GwcNet_GC(d):
+    return GwcNet(d, use_concat_volume=True)
--- a/models/loss.py
+++ b/models/loss.py
@ -0,0 +1,9 @@
+import torch.nn.functional as F
+
+
+def model_loss(disp_ests, disp_gt, mask):
+    weights = [0.5, 0.5, 0.7, 1.0]
+    all_losses = []
+    for disp_est, weight in zip(disp_ests, weights):
+        all_losses.append(weight * F.smooth_l1_loss(disp_est[mask], disp_gt[mask], size_average=True))
+    return sum(all_losses)
--- a/models/submodule.py
+++ b/models/submodule.py
@ -0,0 +1,90 @@
+from __future__ import print_function
+import torch
+import torch.nn as nn
+import torch.utils.data
+from torch.autograd import Variable
+from torch.autograd.function import Function
+import torch.nn.functional as F
+import math
+import numpy as np
+
+
+def convbn(in_channels, out_channels, kernel_size, stride, pad, dilation):
+    return nn.Sequential(nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride,
+                                   padding=dilation if dilation > 1 else pad, dilation=dilation, bias=False),
+                         nn.BatchNorm2d(out_channels))
+
+
+def convbn_3d(in_channels, out_channels, kernel_size, stride, pad):
+    return nn.Sequential(nn.Conv3d(in_channels, out_channels, kernel_size=kernel_size, stride=stride,
+                                   padding=pad, bias=False),
+                         nn.BatchNorm3d(out_channels))
+
+
+def disparity_regression(x, maxdisp):
+    assert len(x.shape) == 4
+    disp_values = torch.arange(0, maxdisp, dtype=x.dtype, device=x.device)
+    disp_values = disp_values.view(1, maxdisp, 1, 1)
+    return torch.sum(x * disp_values, 1, keepdim=False)
+
+
+def build_concat_volume(refimg_fea, targetimg_fea, maxdisp):
+    B, C, H, W = refimg_fea.shape
+    volume = refimg_fea.new_zeros([B, 2 * C, maxdisp, H, W])
+    for i in range(maxdisp):
+        if i > 0:
+            volume[:, :C, i, :, i:] = refimg_fea[:, :, :, i:]
+            volume[:, C:, i, :, i:] = targetimg_fea[:, :, :, :-i]
+        else:
+            volume[:, :C, i, :, :] = refimg_fea
+            volume[:, C:, i, :, :] = targetimg_fea
+    volume = volume.contiguous()
+    return volume
+
+
+def groupwise_correlation(fea1, fea2, num_groups):
+    B, C, H, W = fea1.shape
+    assert C % num_groups == 0
+    channels_per_group = C // num_groups
+    cost = (fea1 * fea2).view([B, num_groups, channels_per_group, H, W]).mean(dim=2)
+    assert cost.shape == (B, num_groups, H, W)
+    return cost
+
+
+def build_gwc_volume(refimg_fea, targetimg_fea, maxdisp, num_groups):
+    B, C, H, W = refimg_fea.shape
+    volume = refimg_fea.new_zeros([B, num_groups, maxdisp, H, W])
+    for i in range(maxdisp):
+        if i > 0:
+            volume[:, :, i, :, i:] = groupwise_correlation(refimg_fea[:, :, :, i:], targetimg_fea[:, :, :, :-i],
+                                                           num_groups)
+        else:
+            volume[:, :, i, :, :] = groupwise_correlation(refimg_fea, targetimg_fea, num_groups)
+    volume = volume.contiguous()
+    return volume
+
+
+class BasicBlock(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride, downsample, pad, dilation):
+        super(BasicBlock, self).__init__()
+
+        self.conv1 = nn.Sequential(convbn(inplanes, planes, 3, stride, pad, dilation),
+                                   nn.ReLU(inplace=True))
+
+        self.conv2 = convbn(planes, planes, 3, 1, pad, dilation)
+
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        out = self.conv1(x)
+        out = self.conv2(out)
+
+        if self.downsample is not None:
+            x = self.downsample(x)
+
+        out += x
+
+        return out
--- a/sceneflow.sh
+++ b/sceneflow.sh
@ -0,0 +1,7 @@
+#!/usr/bin/env bash
+set -x
+DATAPATH="/home/xyguo/data/scene_flow/"
+python main.py --dataset sceneflow \
+    --datapath $DATAPATH --trainlist ./filenames/sceneflow_train.txt --testlist ./filenames/sceneflow_test.txt \
+    --epochs 16 --lrepochs "10,12,14,16:2" \
+    --model gwcnet-gc --logdir ./checkpoints/sceneflow/gwcnet-gc
--- a/utils/init.py
+++ b/utils/init.py
@ -0,0 +1,3 @@
+from utils.experiment import *
+from utils.visualization import *
+from utils.metrics import D1_metric, Thres_metric, EPE_metric
--- a/utils/experiment.py
+++ b/utils/experiment.py
@ -0,0 +1,157 @@
+from __future__ import print_function, division
+import torch
+import torch.nn as nn
+import torch.nn.parallel
+import torch.utils.data
+from torch.autograd import Variable
+import torchvision.utils as vutils
+import torch.nn.functional as F
+import numpy as np
+import time
+from datasets import *
+from models import *
+import copy
+import yaml
+import sys
+import argparse
+
+
+def make_iterative_func(func):
+    def wrapper(vars):
+        if isinstance(vars, list):
+            return [wrapper(x) for x in vars]
+        elif isinstance(vars, tuple):
+            return tuple([wrapper(x) for x in vars])
+        elif isinstance(vars, dict):
+            return {k: wrapper(v) for k, v in vars.items()}
+        else:
+            return func(vars)
+
+    return wrapper
+
+
+def make_nograd_func(func):
+    def wrapper(*f_args, **f_kwargs):
+        with torch.no_grad():
+            ret = func(*f_args, **f_kwargs)
+        return ret
+
+    return wrapper
+
+
+@make_iterative_func
+def tensor2float(vars):
+    if isinstance(vars, float):
+        return vars
+    elif isinstance(vars, torch.Tensor):
+        return vars.data.item()
+    else:
+        raise NotImplementedError("invalid input type for tensor2float")
+
+
+@make_iterative_func
+def tensor2numpy(vars):
+    if isinstance(vars, np.ndarray):
+        return vars
+    elif isinstance(vars, torch.Tensor):
+        return vars.data.cpu().numpy()
+    else:
+        raise NotImplementedError("invalid input type for tensor2numpy")
+
+
+@make_iterative_func
+def check_allfloat(vars):
+    assert isinstance(vars, float)
+
+
+def save_scalars(logger, mode_tag, scalar_dict, global_step):
+    scalar_dict = tensor2float(scalar_dict)
+    for tag, values in scalar_dict.items():
+        if not isinstance(values, list) and not isinstance(values, tuple):
+            values = [values]
+        for idx, value in enumerate(values):
+            scalar_name = '{}/{}'.format(mode_tag, tag)
+            # if len(values) > 1:
+            scalar_name = scalar_name + "_" + str(idx)
+            logger.add_scalar(scalar_name, value, global_step)
+
+
+def save_images(logger, mode_tag, images_dict, global_step):
+    images_dict = tensor2numpy(images_dict)
+    for tag, values in images_dict.items():
+        if not isinstance(values, list) and not isinstance(values, tuple):
+            values = [values]
+        for idx, value in enumerate(values):
+            if len(value.shape) == 3:
+                value = value[:, np.newaxis, :, :]
+            value = value[:1]
+            value = torch.from_numpy(value)
+
+            image_name = '{}/{}'.format(mode_tag, tag)
+            if len(values) > 1:
+                image_name = image_name + "_" + str(idx)
+            logger.add_image(image_name, vutils.make_grid(value, padding=0, nrow=1, normalize=True, scale_each=True),
+                             global_step)
+
+
+def adjust_learning_rate(optimizer, epoch, base_lr, lrepochs):
+    splits = lrepochs.split(':')
+    assert len(splits) == 2
+
+    # parse the epochs to downscale the learning rate (before :)
+    downscale_epochs = [int(eid_str) for eid_str in splits[0].split(',')]
+    # parse downscale rate (after :)
+    downscale_rate = float(splits[1])
+    print("downscale epochs: {}, downscale rate: {}".format(downscale_epochs, downscale_rate))
+
+    lr = base_lr
+    for eid in downscale_epochs:
+        if epoch >= eid:
+            lr /= downscale_rate
+        else:
+            break
+    print("setting learning rate to {}".format(lr))
+    for param_group in optimizer.param_groups:
+        param_group['lr'] = lr
+
+
+class AverageMeter(object):
+    def __init__(self):
+        self.sum_value = 0.
+        self.count = 0
+
+    def update(self, x):
+        check_allfloat(x)
+        self.sum_value += x
+        self.count += 1
+
+    def mean(self):
+        return self.sum_value / self.count
+
+
+class AverageMeterDict(object):
+    def __init__(self):
+        self.data = None
+        self.count = 0
+
+    def update(self, x):
+        check_allfloat(x)
+        self.count += 1
+        if self.data is None:
+            self.data = copy.deepcopy(x)
+        else:
+            for k1, v1 in x.items():
+                if isinstance(v1, float):
+                    self.data[k1] += v1
+                elif isinstance(v1, tuple) or isinstance(v1, list):
+                    for idx, v2 in enumerate(v1):
+                        self.data[k1][idx] += v2
+                else:
+                    assert NotImplementedError("error input type for update AvgMeterDict")
+
+    def mean(self):
+        @make_iterative_func
+        def get_mean(v):
+            return v / float(self.count)
+
+        return get_mean(self.data)
--- a/utils/metrics.py
+++ b/utils/metrics.py
@ -0,0 +1,65 @@
+import torch
+import torch.nn.functional as F
+from utils.experiment import make_nograd_func
+from torch.autograd import Variable
+from torch import Tensor
+
+
+# Update D1 from >3px to >=3px & >5%
+# matlab code:
+# E = abs(D_gt - D_est);
+# n_err = length(find(D_gt > 0 & E > tau(1) & E. / abs(D_gt) > tau(2)));
+# n_total = length(find(D_gt > 0));
+# d_err = n_err / n_total;
+
+def check_shape_for_metric_computation(*vars):
+    assert isinstance(vars, tuple)
+    for var in vars:
+        assert len(var.size()) == 3
+        assert var.size() == vars[0].size()
+
+# a wrapper to compute metrics for each image individually
+def compute_metric_for_each_image(metric_func):
+    def wrapper(D_ests, D_gts, masks, *nargs):
+        check_shape_for_metric_computation(D_ests, D_gts, masks)
+        bn = D_gts.shape[0]  # batch size
+        results = []  # a list to store results for each image
+        # compute result one by one
+        for idx in range(bn):
+            # if tensor, then pick idx, else pass the same value
+            cur_nargs = [x[idx] if isinstance(x, (Tensor, Variable)) else x for x in nargs]
+            if masks[idx].float().mean() / (D_gts[idx] > 0).float().mean() < 0.1:
+                print("masks[idx].float().mean() too small, skip")
+            else:
+                ret = metric_func(D_ests[idx], D_gts[idx], masks[idx], *cur_nargs)
+                results.append(ret)
+        if len(results) == 0:
+            print("masks[idx].float().mean() too small for all images in this batch, return 0")
+            return torch.tensor(0, dtype=torch.float32, device=D_gts.device)
+        else:
+            return torch.stack(results).mean()
+    return wrapper
+
+@make_nograd_func
+@compute_metric_for_each_image
+def D1_metric(D_est, D_gt, mask):
+    D_est, D_gt = D_est[mask], D_gt[mask]
+    E = torch.abs(D_gt - D_est)
+    err_mask = (E > 3) & (E / D_gt.abs() > 0.05)
+    return torch.mean(err_mask.float())
+
+@make_nograd_func
+@compute_metric_for_each_image
+def Thres_metric(D_est, D_gt, mask, thres):
+    assert isinstance(thres, (int, float))
+    D_est, D_gt = D_est[mask], D_gt[mask]
+    E = torch.abs(D_gt - D_est)
+    err_mask = E > thres
+    return torch.mean(err_mask.float())
+
+# NOTE: please do not use this to build up training loss
+@make_nograd_func
+@compute_metric_for_each_image
+def EPE_metric(D_est, D_gt, mask):
+    D_est, D_gt = D_est[mask], D_gt[mask]
+    return F.l1_loss(D_est, D_gt, size_average=True)
--- a/utils/visualization.py
+++ b/utils/visualization.py
@ -0,0 +1,62 @@
+from __future__ import print_function
+import torch
+import torch.nn as nn
+import torch.utils.data
+from torch.autograd import Variable, Function
+import torch.nn.functional as F
+import math
+import numpy as np
+import cv2
+
+# disable multi-thread
+cv2.setNumThreads(0)
+
+
+def gen_error_colormap():
+    cols = np.array(
+        [[0 / 3.0, 0.1875 / 3.0, 49, 54, 149],
+         [0.1875 / 3.0, 0.375 / 3.0, 69, 117, 180],
+         [0.375 / 3.0, 0.75 / 3.0, 116, 173, 209],
+         [0.75 / 3.0, 1.5 / 3.0, 171, 217, 233],
+         [1.5 / 3.0, 3 / 3.0, 224, 243, 248],
+         [3 / 3.0, 6 / 3.0, 254, 224, 144],
+         [6 / 3.0, 12 / 3.0, 253, 174, 97],
+         [12 / 3.0, 24 / 3.0, 244, 109, 67],
+         [24 / 3.0, 48 / 3.0, 215, 48, 39],
+         [48 / 3.0, np.inf, 165, 0, 38]], dtype=np.float32)
+    cols[:, 2: 5] /= 255.
+    return cols
+
+
+error_colormap = gen_error_colormap()
+
+
+class disp_error_image_func(Function):
+    def forward(self, D_est_tensor, D_gt_tensor, abs_thres=3., rel_thres=0.05, dilate_radius=1):
+        D_gt_np = D_gt_tensor.detach().cpu().numpy()
+        D_est_np = D_est_tensor.detach().cpu().numpy()
+        B, H, W = D_gt_np.shape
+        # valid mask
+        mask = D_gt_np > 0
+        # error in percentage. When error <= 1, the pixel is valid since <= 3px & 5%
+        error = np.abs(D_gt_np - D_est_np)
+        error[np.logical_not(mask)] = 0
+        error[mask] = np.minimum(error[mask] / abs_thres, (error[mask] / D_gt_np[mask]) / rel_thres)
+        # get colormap
+        cols = error_colormap
+        # create error image
+        error_image = np.zeros([B, H, W, 3], dtype=np.float32)
+        for i in range(cols.shape[0]):
+            error_image[np.logical_and(error >= cols[i][0], error < cols[i][1])] = cols[i, 2:]
+        # TODO: imdilate
+        # error_image = cv2.imdilate(D_err, strel('disk', dilate_radius));
+        error_image[np.logical_not(mask)] = 0.
+        # show color tag in the top-left cornor of the image
+        for i in range(cols.shape[0]):
+            distance = 20
+            error_image[:, :10, i * distance:(i + 1) * distance, :] = cols[i, 2:]
+
+        return torch.from_numpy(np.ascontiguousarray(error_image.transpose([0, 3, 1, 2])))
+
+    def backward(self, grad_output):
+        return None