search

cavalleria / cavaface

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

What is the configuration/specs of system that you use for training ? #15

Closed pawan9 closed 4 years ago

pawan9 commented 4 years ago

Can you please tell what are the specs of the system that you use for training models and also how time does it take to train a fresh MobileFaceNet, IR-SE-100 model on ms1m-retinanet dataset ?

cavalleria commented 4 years ago

Thx for your question. It takes about 8 hours to train mobilefacenet 24 epoch with 8 V100 GPUS, and 27 hours for IR-SE-100.

pawan9 commented 4 years ago

@cavalleria thanks i really appreciate your work and your quick response

one more question, what is your Training Loss, Training Prec@1 and Training Prec@5 on ms1m-retinanet dataset for MobileFaceNet and IR-SE-100 after 24 epoches ?

cavalleria commented 4 years ago

thx, i check the training logs. For MobileFaceNet, Training Loss 5.6511 (5.7445) Training Prec@1 48.438 (44.848) Training Prec@5 58.008 (56.868) For IR-SE-100, Training Loss 1.9620 (0.9541) Training Prec@1 84.375 (88.590) Training Prec@5 89.062 (93.178) By the way, the validation set acc and megface/ijbc test acc is more important and informative