search

mhamilton723 / STEGO

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

Training on own data without labels #8

Closed IdoZach closed 2 years ago

IdoZach commented 2 years ago

Hi, I would like to train on my own data, without any labels. When running precompute_knns.py I keep running into issues related to missing images or image formats for the labels, which I do not have. What should I do? Thanks.

mhamilton723 commented 2 years ago

Hey @IdoZach, what is your error message i will try to repro

evolu8 commented 2 years ago

If training without labels I found not creating a labels folder was the way forward. The instructions did not make this super clear. Might be worth a note.

mhamilton723 commented 2 years ago

Thank you for that tip @evolu8, i have updated the docs