search

GeWu-Lab / MMCosine_ICASSP23

The code repo for ICASSP 2023 Paper "MMCosine: Multi-Modal Cosine Loss Towards Balanced Audio-Visual Fine-Grained Learning"
18 stars 1 forks source link

What NCE loss is used in your code? #3

Open seongq opened 1 year ago

seongq commented 1 year ago

In "main_CD.py", you implemented NCE loss. Did you use NCE loss in training? If you use NCE loss, which args.lam did you choose?

Rick-Xu315 commented 1 year ago

Hi, Thanks for your interest! NCE loss is implemented only for comparable methods CCL in supplements section 4 as a regularization to CE loss. I tried args.lam from 0.1-1.0 and used the best result as shown in the supplements.