diff --git a/datasets/kitti_dataset.py b/datasets/kitti_dataset.py
index a174a98..3349c60 100644
--- a/datasets/kitti_dataset.py
+++ b/datasets/kitti_dataset.py
@@ -96,4 +96,6 @@ class KITTIDataset(Dataset):
                 return {"left": left_img,
                         "right": right_img,
                         "top_pad": top_pad,
-                        "right_pad": right_pad}
+                        "right_pad": right_pad,
+                        "left_filename": self.left_filenames[index],
+                        "right_filename": self.right_filenames[index]}
diff --git a/kitti12_save.sh b/kitti12_save.sh
new file mode 100755
index 0000000..f122ffd
--- /dev/null
+++ b/kitti12_save.sh
@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -x
+DATAPATH="/home/xyguo/data/kitti_2012/"
+python save_disp.py --datapath $DATAPATH --testlist ./filenames/kitti12_test.txt --model gwcnet-gc --loadckpt ./checkpoints/kitti12/gwcnet-gc/best.ckpt
diff --git a/kitti15_save.sh b/kitti15_save.sh
new file mode 100755
index 0000000..1b78c0d
--- /dev/null
+++ b/kitti15_save.sh
@@ -0,0 +1,4 @@
+#!/usr/bin/env bash
+set -x
+DATAPATH="/home/xyguo/data/kitti_2015/"
+python save_disp.py --datapath $DATAPATH --testlist ./filenames/kitti15_test.txt --model gwcnet-g --loadckpt ./checkpoints/kitti15/gwcnet-g/best.ckpt
diff --git a/save_disp.py b/save_disp.py
new file mode 100644
index 0000000..7daf836
--- /dev/null
+++ b/save_disp.py
@@ -0,0 +1,82 @@
+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 utils import *
+from torch.utils.data import DataLoader
+import gc
+import skimage
+
+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('--maxdisp', type=int, default=192, help='maximum disparity')
+
+parser.add_argument('--dataset', default='kitti', help='dataset name', choices=__datasets__.keys())
+parser.add_argument('--datapath', required=True, help='data path')
+parser.add_argument('--testlist', required=True, help='testing list')
+parser.add_argument('--loadckpt', required=True, help='load the weights from a specific checkpoint')
+
+# parse arguments
+args = parser.parse_args()
+
+# dataset, dataloader
+StereoDataset = __datasets__[args.dataset]
+test_dataset = StereoDataset(args.datapath, args.testlist, False)
+TestImgLoader = DataLoader(test_dataset, 1, shuffle=False, num_workers=4, drop_last=False)
+
+# model, optimizer
+model = __models__[args.model](args.maxdisp)
+model = nn.DataParallel(model)
+model.cuda()
+
+# load parameters
+print("loading model {}".format(args.loadckpt))
+state_dict = torch.load(args.loadckpt)
+model.load_state_dict(state_dict['model'])
+
+
+def test():
+    os.makedirs('./predictions', exist_ok=True)
+    for batch_idx, sample in enumerate(TestImgLoader):
+        start_time = time.time()
+        disp_est_np = tensor2numpy(test_sample(sample))
+        top_pad_np = tensor2numpy(sample["top_pad"])
+        right_pad_np = tensor2numpy(sample["right_pad"])
+        left_filenames = sample["left_filename"]
+        print('Iter {}/{}, time = {:3f}'.format(batch_idx, len(TestImgLoader),
+                                                time.time() - start_time))
+
+        for disp_est, top_pad, right_pad, fn in zip(disp_est_np, top_pad_np, right_pad_np, left_filenames):
+            assert len(disp_est.shape) == 2
+            disp_est = np.array(disp_est[top_pad:, :-right_pad], dtype=np.float32)
+            fn = os.path.join("predictions", fn.split('/')[-1])
+            print("saving to", fn, disp_est.shape)
+            disp_est_uint = np.round(disp_est * 256).astype(np.uint16)
+            skimage.io.imsave(fn, disp_est_uint)
+
+
+# test one sample
+@make_nograd_func
+def test_sample(sample):
+    model.eval()
+    disp_ests = model(sample['left'].cuda(), sample['right'].cuda())
+    return disp_ests[-1]
+
+
+if __name__ == '__main__':
+    test()