search

lorenmt / reco

The implementation of "Bootstrapping Semantic Segmentation with Regional Contrast" [ICLR 2022].
https://shikun.io/projects/regional-contrast
Other
162 stars 25 forks source link

Didn't get the same results as you #17

Closed Haiduongcable closed 2 years ago

Haiduongcable commented 2 years ago

I retrain your model in Pascal Voc in full label setting with semi supervised + reco loss + class mix with 600 labels, but I don't get 72.75% mIOU in val. My result is about 69.5%-70% mIOU in val. Can you give me your pretrained or your training code is right?

lorenmt commented 2 years ago

Have you changed any hyper-parameters?

My three random seeds gave 72.88, 72.60, and 72.78 respectively.

lorenmt commented 2 years ago

I am closing this issue since no further updates are available.