frank-xwang / debiased-pseudo-labeling

[CVPR 2022] Pytorch implementation for “Debiased Learning from Naturally Imbalanced Pseudo-Labels”

Home Page:https://arxiv.org/pdf/2201.01490.pdf

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Bug in modifying number of classes

sandeepmukh opened this issue · comments

In main_DebiasPL.py, qhat is initialized with a hard-coded number of classes (1000), but this should inherit from args.cls. The two locations where I've identified this issue are in initializing the FixMatch model and in intializing qhat.

This shouldn't affect any of the results as Imagenet, but for people adapting the code to new datasets it might create an issue.