search

PRIS-CV / Fine-Grained-or-Not

Code release for Your “Flamingo” is My “Bird”: Fine-Grained, or Not (CVPR 2021 Oral)
MIT License
58 stars 9 forks source link

about convergence speed of training process #2

Closed lynlindasy closed 2 years ago

lynlindasy commented 3 years ago

Hello, I tried your method on CUB Bird but found the printed loss is different from your log in this repo. Is there any details have been changed? Thank you.

dongliangchang commented 2 years ago

no

shenglih commented 2 years ago

hi I tried the stanford car example, the gradient kept exploding... (train_loss = nan, test_loss = nan)... i tried some fixes but none worked... Do you happen to know why by any chance? Thanks so much

dongliangchang commented 2 years ago

@shenglih

  1. Please check the version of the PyTorch. PyTorch 2 and 3 have big differences in the loss function
  2. Please check the folder name, the folder name should be 001,002,003,... instead of 1,2,3,...
shenglih commented 2 years ago

Thank you so much for your prompt reply!!! Changing the folder name worked!! LOL

shenglih commented 2 years ago

Sorry to bug again... but it weirdly worked once but whenever I restarted it remained nan again.... I used python 3.6, torch 1.2.0 and torchvision 0.4.0... Any other clues by any chance? Thanks so much Dongliang!

----UPDATE: ALL GOOD I GOT IT TO WORK!! THANK YOU SO MUCH