Person_reID_baseline_pytorch icon indicating copy to clipboard operation
Person_reID_baseline_pytorch copied to clipboard

Published 20 hours ago • verified publisher icon layumi

how many epochs to reach top performence on market1501?

Open bbbxixixixi opened this issue 4 years ago • 2 comments

I use this command to run train.py, other parameters remain unchanged python train.py --gpu_ids 0 --name ft_ResNet50 --batchsize 32 --data_dir F:\data\market1501\pytorch --erasing_p 0.5 --PCB after 20 epochs, the loss is still very high, from orignial 38 to 37-38, and precision is around 3% in training. Are they the right hyperparameters? like lr 0.05

bbbxixixixi avatar Sep 30 '20 09:09 bbbxixixixi

after 60 epoch, I got result Rank@1:0.290974 Rank@5:0.463777 Rank@10:0.544834 mAP:0.113355

bbbxixixixi avatar Sep 30 '20 13:09 bbbxixixixi

Hi @bbbxixixixi It is weird. Could you check your Numpy Version first? Is the result correct on the baseline model without PCB?

layumi avatar Oct 16 '20 23:10 layumi