Bump up the timm version in the requirements.txt to the latest version of timm.

[sugatoray]

Currently the version of timm for this library is 0.3.2, but the latest version of timm is 0.4.9. When I try installing timm and segmentation-models-pytorch using conda, since the latter can only be installed using pip, this effectively lowers the version of timm to 0.3.2.

I assume that timm follows backward-compatibility and hence, segmentation-models-pytorch should work with later versions of timm as well.

So, is it possible to change the requirements from timm==0.3.2 to timm==0.4.9?

# requirements.txt
torchvision>=0.3.0
pretrainedmodels==0.7.4
efficientnet-pytorch==0.6.3
timm==0.3.2

[psinger]

@qubvel

Would it be possible to set it to

timm >= 0.4.12

Timm should be backwards compatible, and currently it breaks with any newer version of timm due to dependency conflicts.

[qubvel]

@psinger timm developing very fast, I am not sure it would be always backward compatible.
It should be possible to force install a newer version

[psinger]

Problem is that if you put segmentation models in a requirements.txt and then try to install newer timm version it will fail due to dependency restrictions. But I see your point.