Closed kennyvoo closed 1 year ago
Hi, thanks for your interest in our work. I think when training with ViT backbone implemented by TransReID, maybe you should use the original optimizer (sgd). For some references on training with ViT: https://github.com/damo-cv/TransReID-SSL/tree/main/cluster-contrast-reid https://github.com/RikoLi/WACV23-workshop-TMGF
You are right! The result of the training with tansreid_base
Mean AP: 56.7%
CMC Scores:
top-1 83.1%
top-5 89.5%
top-10 91.7%
Hello, great work! I've tried using a stronger backbone transreid_base ICS to train the model while keeping the other parameters the same but the performance is not better than resnet50. I have attached the training logs. Can you provide some advice on this? vit_base_train_log.txt