search

mhamilton723 / STEGO

Unsupervised Semantic Segmentation by Distilling Feature Correspondences
MIT License
711 stars 142 forks source link

Loss linear is None #68

Closed Holmes2002 closed 1 year ago

Holmes2002 commented 1 year ago

Hi, I have trained with my dataset. I trained with 20k images and everything was fine but I'm training with 70k images and linear losss are NONE. Why ?

Holmes2002 commented 1 year ago

Oh, I see . I trained with labels so loss linear is not None. If we train without labels, loss linear will be None, right ?

mhamilton723 commented 1 year ago

Yes @Holmes2002 the linear probe needs labels to train so if your dataset doesn't include labels it won't train those pieces.