有关shape的runtimeerror

Question

关于不同尺寸的输入，导致的报错问题，想问下可能是由什么造成的，该怎么解决。
RuntimeError: shape '[1, 96, 13, 7, 13, 7]' is invalid for input of size 884736
更详细的见下面说明

当输入为224*224时，正常输出

input = torch.rand(1, 3, 224, 224)
print(m1.max_vit_base_224().forward(input)[0].shape)
print(m1.max_vit_base_224().forward(input)[1].shape)
print(m1.max_vit_base_224().forward(input)[2].shape)
print(m1.max_vit_base_224().forward(input)[3].shape)

#结果为：
#torch.Size([1, 96, 56, 56])
#torch.Size([1, 192, 28, 28])
#torch.Size([1, 384, 14, 14])
#torch.Size([1, 768, 7, 7])

但是当输入换成别的尺寸，比如384*384，就会报如下错误：

Traceback (most recent call last):
  File "G:/swinmask/Swin-Transformer-Object-Detection/somettttrrrryy.py", line 10, in <module>
    print(m1.max_vit_base_224().forward(input)[0].shape)
  File "G:\swinmask\Swin-Transformer-Object-Detection\mmdet\models\backbones\maxvit.py", line 688, in forward
    output = stage(output)
  File "D:\Program\anaconda3\envs\swin_det\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "G:\swinmask\Swin-Transformer-Object-Detection\mmdet\models\backbones\maxvit.py", line 560, in forward
    output = self.blocks(input)
  File "D:\Program\anaconda3\envs\swin_det\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "D:\Program\anaconda3\envs\swin_det\lib\site-packages\torch\nn\modules\container.py", line 119, in forward
    input = module(input)
  File "D:\Program\anaconda3\envs\swin_det\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "G:\swinmask\Swin-Transformer-Object-Detection\mmdet\models\backbones\maxvit.py", line 492, in forward
    output = self.grid_transformer(self.block_transformer(self.mb_conv(input)))
  File "D:\Program\anaconda3\envs\swin_det\lib\site-packages\torch\nn\modules\module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "G:\swinmask\Swin-Transformer-Object-Detection\mmdet\models\backbones\maxvit.py", line 399, in forward
    input_partitioned = self.partition_function(input, self.grid_window_size)
  File "G:\swinmask\Swin-Transformer-Object-Detection\mmdet\models\backbones\maxvit.py", line 133, in window_partition
    windows = input.view(B, C, H // window_size[0], window_size[0], W // window_size[1], window_size[1])
RuntimeError: shape '[1, 96, 13, 7, 13, 7]' is invalid for input of size 884736

Answer 1

和你做window划分时的操作有关，私信你了，有不懂的赶紧问