insightface/recognition/partial_fc

Partial-FC

Partial FC is a distributed deep learning training framework for face recognition. The goal of Partial FC is to make training large scale classification task (eg. 10 or 100 millions identies) fast and easy. It is faster than model parallel and can train more identities.

Contents

Partial FC

• Largest Face Recognition Dataset: Glint360k
• Distributed Training Performance
• Citation

Glint360k

We clean, merge, and release the largest and cleanest face recognition dataset Glint360k. Baseline models trained on Glint360k with our proposed training strategy can easily achieve state-of-the-art. The released dataset contains 18 million images of 360K individuals. The performance of Glint360k eval on large-scale test set IFRT, IJB-C and Megaface are as follows:

Evaluation on IFRT

r means the negative class centers sampling rate.

Backbone	Dataset	African	Caucasian	Indian	Asian	ALL
R50	MS1M-V3	76.24	86.21	84.44	37.43	71.02
R124	MS1M-V3	81.08	89.06	87.53	38.40	74.76
R100	Glint360k(r=1.0)	89.50	94.23	93.54	65.07	88.67
R100	Glint360k(r=0.1)	90.45	94.60	93.96	63.91	88.23

Evaluation on IJB-C and Megaface

Our backbone is ResNet100, we set feature scale s to 64 and cosine margin m of CosFace at 0.4. TAR@FAR=1e-4 is reported on the IJB-C datasets, TAR@FAR=1e-6 is reported on Megaface verification.

Test Dataset	IJB-C	Megaface_Id	Megaface_Ver
MS1MV2	96.4	98.3	98.6
Glint360k	97.3	99.1	99.1

Performance

We remove the influence of IO, all experiments use mixed precision training, backbone is ResNet50.

1 Million Identities On 8 RTX2080Ti

Method	GPUs	BatchSize	Memory/M	Throughput img/sec	W
Model Parallel	8	1024	10408	2390	GPU
Partial FC(Ours)	8	1024	8100	2780	GPU

10 Million Identities On 64 RTX2080Ti

Method	GPUs	BatchSize	Memory/M	Throughput img/sec	W
Model Parallel	64	2048	9684	4483	GPU
Partial FC(Ours)	64	4096	6722	12600	GPU

Citation

Coming soon.