syp2ysy/VRP-SAM

Result is not fixed

Opened this issue · 1 comments

YasuoFly commented

Thank you for your great work!
I used Resnet50 as the backbone run on coco2014 dataset fold0 and the prompt type is "scribble".

  1. I run twice, the results are 43.68 and 44.53. I see official code have utils.fix_randseed(args.seed) to fix seed, but i get different results which looks relative large gap.
  2. Is my result correct?
    Waiting for your reply! thank you so much!
syp2ysy commented

It is recommended that you provide more detailed information, such as the number of GPUs used for the experiments, the batch size settings, and other relevant details.