Closed TranThanh96 closed 3 weeks ago
Hi @TranThanh96 ~
In fact, the model we published on MFR was trained using the WebFace42M dataset. For a detailed analysis of the results, please refer to our paper. It is worth mentioning that to ensure our TopoFR model achieves outstanding generalization performance on MFR, we applied insights and tricks we gathered from our experiences in previous competitions during its training.
@DanJun6737 Can you share some tips? and when will you update source code training with partial fc ?
I found your model on MFR: http://iccv21-mfr.com/#/leaderboard/academic Is that correct? Which dataset did you use for training? Have you tried training with WF42M, and if so, what were your results?