Fresh install. Received this error: "ValueError: Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = torch.Size([128, 3, 224, 224])" about big-sleep

[lucidrains]

@Danscap hmm, that's really weird, what OS are you on?

[Danscap]

@lucidrains I am on windows 10 home. Is that the issue?

Could the issue instead be that the version of this software installed from pip is outdated?

[Danscap]

@lucidrains here is the full output of the error:

---

Imagining "a slice of ham with a human mouth" from the depths of my weights...
iteration: 0%| | 0/1050 [00:00<?, ?it/s]
epochs: 0%| | 0/20 [00:00<?, ?it/s]
Traceback (most recent call last):
File "c:\users\jorge\appdata\local\programs\python\python38\lib\runpy.py", line 193, in _run_module_as_main
return _run_code(code, main_globals, None,
File "c:\users\jorge\appdata\local\programs\python\python38\lib\runpy.py", line 86, in run_code
exec(code, run_globals)
File "C:\Users\Jorge\AppData\Local\Programs\Python\Python38\Scripts\dream.exe_main.py", line 7, in
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\big_sleep\cli.py", line 42, in main
fire.Fire(train)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\fire\core.py", line 141, in Fire
component_trace = _Fire(component, args, parsed_flag_args, context, name)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\fire\core.py", line 466, in _Fire
component, remaining_args = _CallAndUpdateTrace(
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\fire\core.py", line 681, in _CallAndUpdateTrace
component = fn(*varargs, **kwargs)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\big_sleep\cli.py", line 39, in train
imagine()
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\big_sleep\big_sleep.py", line 237, in forward
loss = self.train_step(epoch, i)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\big_sleep\big_sleep.py", line 207, in train_step
losses = self.model(self.encoded_text)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\torch\nn\modules\module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\big_sleep\big_sleep.py", line 130, in forward
into = normalize_image(into)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\torchvision\transforms\transforms.py", line 166, in call
return F.normalize(tensor, self.mean, self.std, self.inplace)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\torchvision\transforms\functional.py", line 193, in normalize
raise ValueError('Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = '
ValueError: Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = torch.Size([128, 3, 224, 224]).

---

[lucidrains]

@Danscap I just did a fresh install and it works fine

maybe it is a windows thing, but i'm not sure

[DrJKL]

@Danscap Could you try updating your pip installations?
https://www.activestate.com/resources/quick-reads/how-to-update-all-python-packages/

[Danscap]

@DrJKL while I did have some outdated packages that were updated with the commands from the article. Trying afterwards gave the same result.

It's possible the way this code is structured, it may only work on Linux?

[DrJKL]

I've been running it exclusively on Windows 10.

[htoyryla]

return F.normalize(tensor, self.mean, self.std, self.inplace)
File "c:\users\jorge\appdata\local\programs\python\python38\lib\site-packages\torchvision\transforms\functional.py", line 193, in normalize
raise ValueError('Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = '
ValueError: Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = torch.Size([128, 3, 224, 224]).

Looks like an older version of torchvision is in use which only accepts 3-dim tensors. Update torchvision.

[Danscap]

@htoyryla what torchvision version do you recommend? I have 0.6.1 according to 'print(torchvision.version)'

[htoyryla]

I have 0.8.2. I guess 0.8.x versions are the ones to use with pytorch 1.7.x . There seems been a lot of development going on torchvision transforms, making them more general (to work with both images and batches of tensors).

I am a bit baffled by the line number in your error message, though, the corresponding line in 0.6.1 should be around 290. This is what led me to think that your version is older.

PS. I see, there is a windows specific branch which explains the line number https://github.com/pytorch/vision/blob/3ff9865c0b424ddc4f7ce7f312aa28e604150fa8/torchvision/transforms/functional.py#L192-L194