diff --git a/main.py b/main.py
index 3144215..16e58bc 100644
--- a/main.py
+++ b/main.py
@@ -119,7 +119,7 @@ def train():
         # testing
         avg_test_scalars = AverageMeterDict()
         for batch_idx, sample in enumerate(TestImgLoader):
-            global_step = len(TrainImgLoader) * epoch_idx + batch_idx
+            global_step = len(TestImgLoader) * 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)
diff --git a/models/gwcnet.py b/models/gwcnet.py
index 3622e1a..923e6bd 100644
--- a/models/gwcnet.py
+++ b/models/gwcnet.py
@@ -71,13 +71,13 @@ class hourglass(nn.Module):
         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),
+        self.conv2 = nn.Sequential(convbn_3d(in_channels * 2, in_channels * 2, 3, 1, 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),
+        self.conv4 = nn.Sequential(convbn_3d(in_channels * 4, in_channels * 4, 3, 1, 1),
                                    nn.ReLU(inplace=True))
 
         self.conv5 = nn.Sequential(