search

HuangYG123 / CurricularFace

CurricularFace(CVPR2020)
MIT License
526 stars 72 forks source link

prepare training data #9

Open gaozhangyang opened 4 years ago

gaozhangyang commented 4 years ago

Could you tell me how to prepare the training data and the 'refined_ms1m.txt'? Thanks.

KavyaBPadmesh commented 4 years ago

Do you think this approach will work when we train using dataset little label noise?

Dnekhaev commented 4 years ago

Could anyone share results of comparison of ArcFace and CurricularFace on some public version of MS1M or any other datasets? In my first experience when I train on my own data, MV-Arc-Softmax with t=1.2 works better than ArcFace, but CurricularFace works worse than ArcFace. Seems that training list refined_ms1m.txt is essential to reproduce the results of the paper and make fair comparison.

kwanking commented 4 years ago

Could you tell me how to prepare the training data and the 'refined_ms1m.txt'? Thanks.

looking forward!

garbo-g commented 2 years ago

Could you tell me how to prepare the training data and the 'refined_ms1m.txt'? Thanks.