search

yulu0724 / SDC-IL

Semantic Drift Compensation for Class-Incremental Learning (CVPR2020)
115 stars 34 forks source link

TripletLoss or TripletLossNoHardMining #19

Open GregoirePetit opened 2 years ago

GregoirePetit commented 2 years ago

Hi Lu, I'm currently trying to reproduce your paper results on CIFAR100 and I was wondering which triplet loss version you used since you provided two implementations in your repository (triplet.py and triplet_no_hard_mining.py). Could you provide more information about it? Thanks in advance, Grégoire

yulu0724 commented 2 years ago

Hi Gregoire, for cifar and imagenet-sub, we use triplet_no_hard_mining as the loss.