search

mhamilton723 / STEGO

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

Add to README Instructions on using demo_segmentation.py? #7

Closed timothyjchurchill closed 1 year ago

timothyjchurchill commented 2 years ago

Hello! Could you add to the README instructions on how to try out a pretrained model on a new image/dataset? Would that be done through modifying demo_config.yml and using demo_segmentation.py? Sorry for the dumb question. Thank you!

maxjeblick commented 2 years ago

A colab example notebook may also be helpful to try things out easily.

alien19 commented 2 years ago

Hi @timothyjchurchill, I have tried the demo_segmentation script by just python demo_segmentation.py form inside STEGO/src and before that I had to edit the path parameters in the demo_config.yml.

Hope you solved it!

mhamilton723 commented 2 years ago

@maxjeblick added a quick colab example, hopefully will be able to build it out more in the next few weeks

maxjeblick commented 2 years ago

Thanks a lot, appreciate it!

Lance4Liu commented 7 months ago

Hi @timothyjchurchill, I have tried the demo_segmentation script by just python demo_segmentation.py form inside STEGO/src and before that I had to edit the path parameters in the demo_config.yml.

Hope you solved it!

@alien19 Hello! I also changed the models path of demo_config.yml (to one of the checkpoints files we saved when we ran train_segment.py) and ran demo_segment.py, but my Results were a blob of black. There is no desired effect. Could you please tell me how to modify demo_config.yml in more detail? For example, what should model_path change? And how many epochs should it run? I ran 50 or so epochs and it didn't work as expected.