search

cavalleria / cavaface

face recognition training project(pytorch)
MIT License
456 stars 88 forks source link

arcface(s=64,m=0.5) Accuracy is worse than Softmax #60

Open zhanglaplace opened 3 years ago

zhanglaplace commented 3 years ago

train data: glint360Datasets(https://github.com/deepinsight/insightface/blob/master/recognition/partial_fc/README.md); 1、 First，train baseline model with R50-Softmax 2、Second，Train ArcFace-Softmax by finetune the backbone using Step1 Backbone model . the acc is lower then origin softmax; image

cavalleria commented 3 years ago

I have not finetuned using arcface, but the result does not seem reasonable.

zhanglaplace commented 3 years ago

I have not finetuned using arcface, but the result does not seem reasonable. train stretch using arcface has a better performance ?

cavalleria commented 3 years ago

training from scratch using arcface is better than softmax. i think properly finetune on softmax with arcface should better softmax