ma-lab-berkeley / mcr2 Goto Github PK

Hi, i have read the paper, and i'm very excited that MCR2 code will generate structrual features, which will be very helpful for many downstream tasks.
However, when i try the CIFAR100-selfsup code from Commands for Self-supervised Learning Setting section, it seems that the learned model is failed(very small weights like 1e-27) and extracted features are all zero. But i am not sure what's the problem with this code.

CIFAR10 is workable, but similar code for CIFAR100 is failed. My code is followed as
python3 train_selfsup.py --arch resnet18ctrl --data cifar100 --fd 128 --epo 100 --bs 1000 --eps 0.5 --gam1 20 --gam2 0.05 --lr 0.1 --aug 50 --transform cifarfrom
python extract.py--model dir .saved models/selfsupresnet18ctr1+128far100 epo100 bs1000 aug50+cifar lr0.1 mom0.9 wd0.0005 gam120.0 gam20.05 eps0.5/

I'm appreciate if you can solve my problem!