How to use image size other than 224

[alqurri77]

Hi
How to use image size other than 224

thanks

[BinYCn]

Thank you for your question.

To use an image size other than 224, you can adjust the WINDOW_SIZE parameter in the configuration file located at
./configs/*.yaml. For example, set WINDOW_SIZE to 16 when the image size is 512.

[alqurri77]

I did that, but the issue I'm getting this error:

my_weights torch.Size([256, 3, 256, 256])
error new_shape torch.Size([256, 256, 768])
  0%|                                         | 0/150 [00:05<?, ?it/s]
---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
Input In [17], in <cell line: 116>()
    148 #net.load_from(config)
    151 trainer = {'Synapse': trainer_synapse,}
--> 152 trainer[dataset_name](args, net, args.output_dir)

Input In [16], in trainer_synapse(args, model, snapshot_path)
     54 image_batch, label_batch = image_batch.to(device), label_batch.to(device)
     55 #d1, d2, d3 = model(image_batch)
     56 
     57 
   (...)
     60 
     61 #loss_ce =  muti_bc_loss_fusion(d1, d2, d3,  label_batch ,ce_loss)
---> 63 ds1,ds2,ds3,d1, d2, d3,d4,d5,d6= model(image_batch)
     65 loss_dice =muti_dice_loss_fusion(ds1,ds2,ds3,d1, d2, d3,d4,d5,d6, label_batch,dice_loss)
     67 loss_ce =  muti_bc_loss_fusion  (ds1,ds2,ds3,d1, d2, d3,d4, d5,d6, label_batch ,ce_loss) #ce_loss(d4, y[:].long())

File /scratch/ahmed/lib/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
   1516     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1517 else:
-> 1518     return self._call_impl(*args, **kwargs)

File /scratch/ahmed/lib/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
   1522 # If we don't have any hooks, we want to skip the rest of the logic in
   1523 # this function, and just call forward.
   1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1525         or _global_backward_pre_hooks or _global_backward_hooks
   1526         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527     return forward_call(*args, **kwargs)
   1529 try:
   1530     result = None

Input In [10], in SwinUnet.forward(self, x)
   1411 if x.size()[1] == 1:
   1412     x = x.repeat(1, 3, 1, 1)
-> 1413 logits = self.swin_unet(x)
   1414 return logits

File /scratch/ahmed/lib/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
   1516     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1517 else:
-> 1518     return self._call_impl(*args, **kwargs)

File /scratch/ahmed/lib/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
   1522 # If we don't have any hooks, we want to skip the rest of the logic in
   1523 # this function, and just call forward.
   1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1525         or _global_backward_pre_hooks or _global_backward_hooks
   1526         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527     return forward_call(*args, **kwargs)
   1529 try:
   1530     result = None

Input In [10], in SwinTransformerSys.forward(self, x)
   1168 output4 = self.final4(x0_4_unet2)
   1170 #--------------------------------------------------------------------------------
-> 1174 x, x_downsample,x_downsample_my_weights = self.forward_features(x)
   1175 U_decoder_thing = []
   1177 #print("encoders_results[0] ",encoders_results[0].shape)
   1178 #print("encoders_results[1] ",encoders_results[1].shape)
   1179 #print("encoders_results[2] ",encoders_results[2].shape)
   1180 #print("encoders_results[3] ",encoders_results[3].shape)

Input In [10], in SwinTransformerSys.forward_features(self, x)
   1036 for layer in self.layers:
   1037     x_downsample.append(x)
-> 1039     x,my_weights = layer(x)
   1040     x_downsample_my_weights.append(my_weights)
   1043 x = self.norm(x)  # B L C

File /scratch/ahmed/lib/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
   1516     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1517 else:
-> 1518     return self._call_impl(*args, **kwargs)

File /scratch/ahmed/lib/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
   1522 # If we don't have any hooks, we want to skip the rest of the logic in
   1523 # this function, and just call forward.
   1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1525         or _global_backward_pre_hooks or _global_backward_hooks
   1526         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527     return forward_call(*args, **kwargs)
   1529 try:
   1530     result = None

Input In [10], in BasicLayer.forward(self, x)
    612         x = checkpoint.checkpoint(blk, x)
    613     else:
--> 614         x,my_weights = blk(x)
    615 if self.downsample is not None:
    616     x = self.downsample(x)

File /scratch/ahmed/lib/torch/nn/modules/module.py:1518, in Module._wrapped_call_impl(self, *args, **kwargs)
   1516     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1517 else:
-> 1518     return self._call_impl(*args, **kwargs)

File /scratch/ahmed/lib/torch/nn/modules/module.py:1527, in Module._call_impl(self, *args, **kwargs)
   1522 # If we don't have any hooks, we want to skip the rest of the logic in
   1523 # this function, and just call forward.
   1524 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1525         or _global_backward_pre_hooks or _global_backward_hooks
   1526         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1527     return forward_call(*args, **kwargs)
   1529 try:
   1530     result = None

Input In [10], in SwinTransformerBlock.forward(self, x)
    401 #elif new_shape2.shape[2]==576:
    402 #  new_shape3=self.chaneshape5(new_shape2)
    403 #elif new_shape2.shape[2]==1024:
    404 #  new_shape3=self.chaneshape6(new_shape2)
    405 else:
    406     print("error new_shape",new_shape2.shape)
--> 407 print("my_weights3:",new_shape3.shape)
    408 my_weights=new_shape3
    409 # merge windows

UnboundLocalError: local variable 'new_shape3' referenced before assignment

I think the problem with this code:

if new_shape2.shape[2]==147:
          new_shape3=self.chaneshape(new_shape2)
        elif new_shape2.shape[2]==294:
          new_shape3=self.chaneshape2(new_shape2)
        elif new_shape2.shape[2]==588:
          new_shape3=self.chaneshape3(new_shape2)
        elif new_shape2.shape[2]==1176:
          new_shape3=self.chaneshape4(new_shape2)
        
        else:
            print("error new_shape",new_shape2.shape)

for 512, the size will be 1024. but there is no 1024 option there.

[BinYCn]

You can set WINDOW_SIZE to 32 when the image size is 1024.

[alqurri77]

Thank you. I meant the code at "SwinTransformerBlock" class, its not working even if you set the size of the window correctly. It stuck on this code fragment:

new_shape2=new_shape.reshape(new_shape.shape[0], new_shape.shape[1], -1)
        
        if new_shape2.shape[2]==147:
          new_shape3=self.chaneshape(new_shape2)
        elif new_shape2.shape[2]==294:
          new_shape3=self.chaneshape2(new_shape2)
        elif new_shape2.shape[2]==588:
          new_shape3=self.chaneshape3(new_shape2)
        elif new_shape2.shape[2]==1176:
          new_shape3=self.chaneshape4(new_shape2)
        
        else:
            print("error new_shape",new_shape2.shape)