Merge pull request #194 from pedropesserl/master

fix issues with training in version 1 and sampling with custom dataset

Author: WuJunde

guided_diffusion/custom_dataset_loader.py

@@ -23,8 +23,8 @@ def __init__(self, args, data_path , transform = None, mode = 'Training',plane =
         images = sorted(glob(os.path.join(path, "images/*.png")))
         masks = sorted(glob(os.path.join(path, "masks/*.png")))
 
-        self.name_list = images[:2]
-        self.label_list = masks[:2]
+        self.name_list = images
+        self.label_list = masks
         self.data_path = path
         self.mode = mode
 
@@ -44,18 +44,19 @@ def __getitem__(self, index):
         img = Image.open(img_path).convert('RGB')
         mask = Image.open(msk_path).convert('L')
 
-        if self.mode == 'Training':
-            label = 0 if self.label_list[index] == 'benign' else 1
-        else:
-            label = int(self.label_list[index])
+        # if self.mode == 'Training':
+        #     label = 0 if self.label_list[index] == 'benign' else 1
+        # else:
+        #     label = int(self.label_list[index])
 
         if self.transform:
             state = torch.get_rng_state()
             img = self.transform(img)
             torch.set_rng_state(state)
             mask = self.transform(mask)
 
-        if self.mode == 'Training':
-            return (img, mask, name)
-        else:
-            return (img, mask, name)
+        return (img, mask, name)
+        # if self.mode == 'Training':
+        #     return (img, mask, name)
+        # else:
+        #     return (img, mask, name)

guided_diffusion/unet.py

@@ -735,6 +735,17 @@ def convert_to_fp32(self):
         self.middle_block.apply(convert_module_to_f32)
         self.output_blocks.apply(convert_module_to_f32)
 
+    def load_part_state_dict(self, state_dict):
+
+        own_state = self.state_dict()
+        for name, param in state_dict.items():
+            if name not in own_state:
+                    continue
+            if isinstance(param, th.nn.Parameter):
+                # backwards compatibility for serialized parameters
+                param = param.data
+            own_state[name].copy_(param)
+
     def enhance(self, c, h):
         cu = layer_norm(c.size()[1:])(c)
         hu = layer_norm(h.size()[1:])(h)

scripts/segmentation_sample.py

@@ -17,6 +17,7 @@
 from guided_diffusion import dist_util, logger
 from guided_diffusion.bratsloader import BRATSDataset, BRATSDataset3D
 from guided_diffusion.isicloader import ISICDataset
+from guided_diffusion.custom_dataset_loader import CustomDataset
 import torchvision.utils as vutils
 from guided_diffusion.utils import staple
 from guided_diffusion.script_util import (
@@ -58,6 +59,13 @@ def main():
 
         ds = BRATSDataset3D(args.data_dir,transform_test)
         args.in_ch = 5
+    else:
+        tran_list = [transforms.Resize((args.image_size,args.image_size)), transforms.ToTensor()]
+        transform_test = transforms.Compose(tran_list)
+
+        ds = CustomDataset(args, args.data_dir, transform_test, mode = 'Test')
+        args.in_ch = 4
+
     datal = th.utils.data.DataLoader(
         ds,
         batch_size=args.batch_size,