mhamilton723 / STEGO

Unsupervised Semantic Segmentation by Distilling Feature Correspondences
MIT License
712 stars 143 forks source link

no module named 'core' #11

Closed Cemm23333 closed 1 year ago

Cemm23333 commented 2 years ago

Hello,Hi, thanks for sharing your code! I met with a question when I ran eval_segmention.py.How to deal with it?Thank you!

image

JXFOnestep commented 2 years ago

Me too! Have you fix it?

MY-LIU100101 commented 2 years ago

In the previous code. The src/utils.py was called src/core.py. I think that is the origin of the problem.

I avoid this error by changing the "run_picie" from True to False in the src/configs/eval_config.yml. (The picie is another unsupervised segmentation method compared in this work)

mhamilton723 commented 2 years ago

@MY-LIU100101 thanks for finding the fix for this issue, I will double check on my side to see whether this is still happening on master

Supgb commented 2 years ago

The inappropriate model-saving approach might cause this issue. https://github.com/pytorch/pytorch/issues/18325#issuecomment-661759359 Saving only the state dict instead of the whole model object is more recommended. https://pytorch.org/docs/master/notes/serialization.html#recommended-approach-for-saving-a-model