zhoudaxia233 / efficientunet-pytorch Goto Github PK

Thank you for your sharing!!! when I run with single GPU，it runs well, but when I run with multi GPU, it occur error
RuntimeError: Function CatBackward returned an invalid gradient at index 1 - expected device cuda:1 but got cuda:0
could you give some advice on this error?

requires pytorch >1.0.0, my version is 1.0.1, i found v1.0.1 not support 'load_state_dict_from_url'
but >=1.1.0

I want to give 4th channel input to the U-net but i am facing problem. Do you have any idea?

Why is the channel number of the block parameter in unet [16,24,40,80] and why not [16,24,40,112]

Storage links for pretrained weights are not working.

Issue solved.

i see that in size dict for every b series model the last item size is 32 which is an input to the last node of Unet part of every efficient net model. Can we change 32 to 64 ?

In your decoder, you fixed channels as 512-256-128....
However in general, the decoder channels should be the same as the encoder channels they concat with.

Since channels in efficientnetb0-b7 have big differences, I want to know why you fixed decoder channels. Is it proved to be better by some experiments?

It's a strange error when i inference model efficientb5 unet with tensor with shape 43128800, seems torch.cat([x, blocks.popitem()[1]], dim=1) error, i print shape as upper image, but it shows only in some batchs, i'm sure input tensor shape is all same (batch3128800) anyone help?

Traceback (most recent call last): File "run.py", line 17, in <module> torch.save(model, 'model.pt') File "/opt/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 224, in save return _with_file_like(f, "wb", lambda f: _save(obj, f, pickle_module, pickle_protocol)) File "/opt/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 149, in _with_file_like return body(f) File "/opt/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 224, in <lambda> return _with_file_like(f, "wb", lambda f: _save(obj, f, pickle_module, pickle_protocol)) File "/opt/anaconda3/lib/python3.7/site-packages/torch/serialization.py", line 297, in _save pickler.dump(obj) AttributeError: Can't pickle local object 'EfficientNet.encoder.<locals>.Encoder'

urllib.error.HTTPError: HTTP Error 403: Forbidden