Merge pull request CSAILVision#64 from CSAILVision/hang-dev

fix some torch0.4 warnings

Author: hangzhaomit

eval.py

@@ -7,7 +7,6 @@
 import numpy as np
 import torch
 import torch.nn as nn
-from torch.autograd import Variable
 from scipy.io import loadmat
 # Our libs
 from dataset import ValDataset
@@ -55,7 +54,6 @@ def evaluate(segmentation_module, loader, args):
         with torch.no_grad():
             segSize = (seg_label.shape[0], seg_label.shape[1])
             pred = torch.zeros(1, args.num_class, segSize[0], segSize[1])
-            pred = Variable(pred).cuda()
 
             for img in img_resized_list:
                 feed_dict = batch_data.copy()
@@ -66,7 +64,7 @@ def evaluate(segmentation_module, loader, args):
 
                 # forward pass
                 pred_tmp = segmentation_module(feed_dict, segSize=segSize)
-                pred = pred + pred_tmp / len(args.imgSize)
+                pred = pred + pred_tmp.cpu() / len(args.imgSize)
 
             _, preds = torch.max(pred.data.cpu(), dim=1)
             preds = as_numpy(preds.squeeze(0))
@@ -101,13 +99,16 @@ def main(args):
 
     # Network Builders
     builder = ModelBuilder()
-    net_encoder = builder.build_encoder(arch=args.arch_encoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_encoder)
-    net_decoder = builder.build_decoder(arch=args.arch_decoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_decoder,
-                                        use_softmax=True)
+    net_encoder = builder.build_encoder(
+        arch=args.arch_encoder,
+        fc_dim=args.fc_dim,
+        weights=args.weights_encoder)
+    net_decoder = builder.build_decoder(
+        arch=args.arch_decoder,
+        fc_dim=args.fc_dim,
+        num_class=args.num_class,
+        weights=args.weights_decoder,
+        use_softmax=True)
 
     crit = nn.NLLLoss(ignore_index=-1)
 
@@ -183,8 +184,6 @@ def main(args):
     args = parser.parse_args()
     print(args)
 
-    # torch.cuda.set_device(args.gpu_id)
-
     # absolute paths of model weights
     args.weights_encoder = os.path.join(args.ckpt, args.id,
                                         'encoder' + args.suffix)

eval_multipro.py

@@ -9,7 +9,6 @@
 import math
 import torch
 import torch.nn as nn
-from torch.autograd import Variable
 from scipy.io import loadmat
 # Our libs
 from dataset import ValDataset
@@ -55,7 +54,6 @@ def evaluate(segmentation_module, loader, args, dev_id, result_queue):
         with torch.no_grad():
             segSize = (seg_label.shape[0], seg_label.shape[1])
             pred = torch.zeros(1, args.num_class, segSize[0], segSize[1])
-            pred = Variable(pred).cuda()
 
             for img in img_resized_list:
                 feed_dict = batch_data.copy()
@@ -66,7 +64,7 @@ def evaluate(segmentation_module, loader, args, dev_id, result_queue):
 
                 # forward pass
                 pred_tmp = segmentation_module(feed_dict, segSize=segSize)
-                pred = pred + pred_tmp / len(args.imgSize)
+                pred = pred + pred_tmp.cpu() / len(args.imgSize)
 
             _, preds = torch.max(pred.data.cpu(), dim=1)
             preds = as_numpy(preds.squeeze(0))
@@ -85,7 +83,7 @@ def evaluate(segmentation_module, loader, args, dev_id, result_queue):
 
 def worker(args, dev_id, start_idx, end_idx, result_queue):
     torch.cuda.set_device(dev_id)
- 
+
     # Dataset and Loader
     dataset_val = ValDataset(
         args.list_val, args, max_sample=args.num_val,
@@ -96,16 +94,19 @@ def worker(args, dev_id, start_idx, end_idx, result_queue):
         shuffle=False,
         collate_fn=user_scattered_collate,
         num_workers=2)
-    
+
     # Network Builders
     builder = ModelBuilder()
-    net_encoder = builder.build_encoder(arch=args.arch_encoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_encoder)
-    net_decoder = builder.build_decoder(arch=args.arch_decoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_decoder,
-                                        use_softmax=True)
+    net_encoder = builder.build_encoder(
+        arch=args.arch_encoder,
+        fc_dim=args.fc_dim,
+        weights=args.weights_encoder)
+    net_decoder = builder.build_decoder(
+        arch=args.arch_decoder,
+        fc_dim=args.fc_dim,
+        num_class=args.num_class,
+        weights=args.weights_decoder,
+        use_softmax=True)
 
     crit = nn.NLLLoss(ignore_index=-1)
 
@@ -132,7 +133,7 @@ def main(args):
     nr_files_per_dev = math.ceil(nr_files / nr_devs)
 
     pbar = tqdm(total=nr_files)
-    
+
     acc_meter = AverageMeter()
     intersection_meter = AverageMeter()
     union_meter = AverageMeter()
@@ -146,7 +147,7 @@ def main(args):
         print('process:%d, start_idx:%d, end_idx:%d' % (dev_id, start_idx, end_idx))
         proc.start()
         procs.append(proc)
-   
+
     # master fetches results
     processed_counter = 0
     while processed_counter < nr_files:
@@ -158,10 +159,10 @@ def main(args):
         union_meter.update(union)
         processed_counter += 1
         pbar.update(1)
-    
+
     for p in procs:
         p.join()
-    
+
     iou = intersection_meter.sum / (union_meter.sum + 1e-10)
     for i, _iou in enumerate(iou):
         print('class [{}], IoU: {}'.format(i, _iou))

models/models.py

@@ -64,7 +64,7 @@ class ModelBuilder():
     def weights_init(self, m):
         classname = m.__class__.__name__
         if classname.find('Conv') != -1:
-            nn.init.kaiming_normal(m.weight.data)
+            nn.init.kaiming_normal_(m.weight.data)
         elif classname.find('BatchNorm') != -1:
             m.weight.data.fill_(1.)
             m.bias.data.fill_(1e-4)
@@ -295,7 +295,8 @@ def forward(self, conv_out, segSize=None):
         x = self.conv_last(x)
 
         if self.use_softmax:  # is True during inference
-            x = nn.functional.upsample(x, size=segSize, mode='bilinear')
+            x = nn.functional.upsample(
+                x, size=segSize, mode='bilinear', align_corners=False)
             x = nn.functional.softmax(x, dim=1)
             return x
 
@@ -327,7 +328,8 @@ def forward(self, conv_out, segSize=None):
         x = self.conv_last(x)
 
         if self.use_softmax: # is True during inference
-            x = nn.functional.upsample(x, size=segSize, mode='bilinear')
+            x = nn.functional.upsample(
+                x, size=segSize, mode='bilinear', align_corners=False)
             x = nn.functional.softmax(x, dim=1)
         else:
             x = nn.functional.log_softmax(x, dim=1)
@@ -370,13 +372,14 @@ def forward(self, conv_out, segSize=None):
             ppm_out.append(nn.functional.upsample(
                 pool_scale(conv5),
                 (input_size[2], input_size[3]),
-                mode='bilinear'))
+                mode='bilinear', align_corners=False))
         ppm_out = torch.cat(ppm_out, 1)
 
         x = self.conv_last(ppm_out)
 
         if self.use_softmax:  # is True during inference
-            x = nn.functional.upsample(x, size=segSize, mode='bilinear')
+            x = nn.functional.upsample(
+                x, size=segSize, mode='bilinear', align_corners=False)
             x = nn.functional.softmax(x, dim=1)
         else:
             x = nn.functional.log_softmax(x, dim=1)
@@ -421,13 +424,14 @@ def forward(self, conv_out, segSize=None):
             ppm_out.append(nn.functional.upsample(
                 pool_scale(conv5),
                 (input_size[2], input_size[3]),
-                mode='bilinear'))
+                mode='bilinear', align_corners=False))
         ppm_out = torch.cat(ppm_out, 1)
 
         x = self.conv_last(ppm_out)
 
         if self.use_softmax:  # is True during inference
-            x = nn.functional.upsample(x, size=segSize, mode='bilinear')
+            x = nn.functional.upsample(
+                x, size=segSize, mode='bilinear', align_corners=False)
             x = nn.functional.softmax(x, dim=1)
             return x
 
@@ -497,7 +501,7 @@ def forward(self, conv_out, segSize=None):
             ppm_out.append(pool_conv(nn.functional.upsample(
                 pool_scale(conv5),
                 (input_size[2], input_size[3]),
-                mode='bilinear')))
+                mode='bilinear', align_corners=False)))
         ppm_out = torch.cat(ppm_out, 1)
         f = self.ppm_last_conv(ppm_out)
 
@@ -506,7 +510,8 @@ def forward(self, conv_out, segSize=None):
             conv_x = conv_out[i]
             conv_x = self.fpn_in[i](conv_x) # lateral branch
 
-            f = nn.functional.upsample(f, size=conv_x.size()[2:], mode='bilinear') # top-down branch
+            f = nn.functional.upsample(
+                f, size=conv_x.size()[2:], mode='bilinear', align_corners=False) # top-down branch
             f = conv_x + f
 
             fpn_feature_list.append(self.fpn_out[i](f))
@@ -518,12 +523,13 @@ def forward(self, conv_out, segSize=None):
             fusion_list.append(nn.functional.upsample(
                 fpn_feature_list[i],
                 output_size,
-                mode='bilinear'))
+                mode='bilinear', align_corners=False))
         fusion_out = torch.cat(fusion_list, 1)
         x = self.conv_last(fusion_out)
 
         if self.use_softmax:  # is True during inference
-            x = nn.functional.upsample(x, size=segSize, mode='bilinear')
+            x = nn.functional.upsample(
+                x, size=segSize, mode='bilinear', align_corners=False)
             x = nn.functional.softmax(x, dim=1)
             return x
 

test.py

@@ -7,7 +7,6 @@
 import numpy as np
 import torch
 import torch.nn as nn
-from torch.autograd import Variable
 from scipy.io import loadmat
 # Our libs
 from dataset import TestDataset
@@ -48,7 +47,6 @@ def test(segmentation_module, loader, args):
 
         with torch.no_grad():
             pred = torch.zeros(1, args.num_class, segSize[0], segSize[1])
-            pred = Variable(pred).cuda()
 
             for img in img_resized_list:
                 feed_dict = batch_data.copy()
@@ -59,9 +57,9 @@ def test(segmentation_module, loader, args):
 
                 # forward pass
                 pred_tmp = segmentation_module(feed_dict, segSize=segSize)
-                pred = pred + pred_tmp / len(args.imgSize)
+                pred = pred + pred_tmp.cpu() / len(args.imgSize)
 
-            _, preds = torch.max(pred.data.cpu(), dim=1)
+            _, preds = torch.max(pred, dim=1)
             preds = as_numpy(preds.squeeze(0))
 
         # visualization
@@ -78,13 +76,16 @@ def main(args):
 
     # Network Builders
     builder = ModelBuilder()
-    net_encoder = builder.build_encoder(arch=args.arch_encoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_encoder)
-    net_decoder = builder.build_decoder(arch=args.arch_decoder,
-                                        fc_dim=args.fc_dim,
-                                        weights=args.weights_decoder,
-                                        use_softmax=True)
+    net_encoder = builder.build_encoder(
+        arch=args.arch_encoder,
+        fc_dim=args.fc_dim,
+        weights=args.weights_encoder)
+    net_decoder = builder.build_decoder(
+        arch=args.arch_decoder,
+        fc_dim=args.fc_dim,
+        num_class=args.num_class,
+        weights=args.weights_decoder,
+        use_softmax=True)
 
     crit = nn.NLLLoss(ignore_index=-1)
 
@@ -157,8 +158,6 @@ def main(args):
     args = parser.parse_args()
     print(args)
 
-    # torch.cuda.set_device(args.gpu_id)
-
     # absolute paths of model weights
     args.weights_encoder = os.path.join(args.model_path,
                                         'encoder' + args.suffix)