MagFace

MagFace: A Universal Representation for Face Recognition and Quality Assessment
in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2021, Oral presentation.

Project Page: https://irvingmeng.github.io/projects/magface/

Paper: arXiv

知乎解读: https://zhuanlan.zhihu.com/p/475775106

A toy example: examples.ipynb

Poster: GoogleDrive, BaiduDrive code: dt9e

Beamer: GoogleDrive, BaiduDrive, code: c16b

Presentation:

1. CVPR 5-minute presentation.
2. Will release a detailed version later.

NOTE: The original codes are implemented on a private codebase and will not be released. This repo is an official but abridged version. See todo list for plans.

BibTex

@inproceedings{meng2021magface,
  title={{MagFace}: A universal representation for face recognition and quality assessment},
  author={Meng, Qiang and Zhao, Shichao and Huang, Zhida and Zhou, Feng},
  booktitle=CVPR,
  year=2021
}

Model Zoo

Parallel Method	Loss	Backbone	Dataset	Split FC?	Model	Log File
DDP	MagFace	iResNet100	MS1MV2	Yes	GoogleDrive, BaiduDrive code: wsw3	Trained by original codes
DDP	MagFace	iResNet50	MS1MV2	Yes	BaiduDrive code: idkx	BaiduDrive, code: 66j1
DDP	Mag-CosFace	iResNet50	MS1MV2	Yes	BaiduDrive code: rg2w	BaiduDrive, code: ejec
DP	MagFace	iResNet50	MS1MV2	No	BaiduDrive code: tvyv	BaiduDrive, code: hpbt
DP	MagFace	iResNet18	CASIA-WebFace	No	BaiduDrive code: fkja	BaiduDrive, code: qv2x
DP	ArcFace	iResNet18	CASIA-WebFace	No	BaiduDrive code: wq2w	BaiduDrive, code: 756e

Evaluation

Face Recognition

Steps to evaluate modes on lfw/cfp/agedb:

1. download data from GDrive or BaiduDrive, code: z7hs
2. cd eval/eval_recognition/ and extract the data in the folder
3. evaluate the model by with eval.sh (e.g., ./eval.sh magface_epoch_00025.pth official 100)

Use eval_ijb.sh for evaluation on IJB-B (Gdrive orBaiduDrive code: iiwa) and IJB-C (Gdrive or BaiduDrive code: q6md). Please apply for permissions from NIST before your usage.

Quality Assessment

Steps to calculate face qualities (examples.ipynb is a toy example).

1. extract features from faces with inference/gen_feat.py.
2. calculate feature magnitudes with np.linalg.norm().

Plot the error-versus-reject curve:

1. prepare the features (in the recognition step).
2. cd eva/eval_quality and run eval_quality.sh (e.g., ./eval_quality.sh lfw).

Note: model used in the quality assessment session of the paper can be found here.

Basic Training

1. install requirements.
2. Align images to 112x112 pixels with 5 facial landmarks (code).
3. Prepare a training list with format imgname 0 id 0 in each line (id starts from 0), as indicated here. In the paper, we employ MS1MV2 as the training dataset which can be downloaded from InsightFace (MS1M-ArcFace in DataZoo). Use rec2image.py to extract images.
4. Modify parameters in run.sh/run_dist.sh/run_dist_cos.sh and run it.

Parallel Training

Note: Use Pytorch > 1.7 for this feature. Codes are mainly based on torchshard from Kaiyu Yue.

How to run:

1. Update NCCL info (can be found with the command ifconfig) and port info in train_dist.py
2. Set the number of gpus in here.
3. [Optional. Not tested yet!] If training with multi-machines, modify node number.
4. [Optional. Help needed as NAN can be reached during training.] Enable fp16 training by setiing --fp16 1 in run/run_dist.sh.
5. run run/run_dist.sh.

Parallel training (Sec. 5.1 in ArcFace) can highly speed up training as well as reduce consumption of GPU memory. Here are some results.

Parallel Method	Float Type	Backbone	GPU	Batch Size	FC Size	Split FC?	Avg. Throughput (images/sec)	Memory (MiB)
DP	FP32	iResNet50	v100 x 8	512	85742	No	1099.41	8681
DDP	FP32	iResNet50	v100 x 8	512	85742	Yes	1687.71	8137
DDP	FP16	iResNet50	v100 x 8	512	85742	Yes	3388.66	5629
DP	FP32	iResNet100	v100 x 8	512	85742	No	612.40	11825
DDP	FP32	iResNet100	v100 x 8	512	85742	Yes	1060.16	10777
DDP	FP16	iResNet100	v100 x 8	512	85742	Yes	2013.90	7319

Tips for MagFace

1. In practical, one may want to finetune a existing model for either performance boosts or quality-aware ability. This is practicable (verified in our scenario) but requires a few modifications. Here are my recommended steps:
   • Generate features from a few samples by existing model and calculate their magnitudes.
   • Assume that magnitudes are distributed in [x1, x2], then modify parameters to meet l_a < x1, u_a > x2.
   • In our scenario, we have a model trained by ArcFace which produces magnitudes around 1. [l_a, u_a, l_m, u_m, l_g] =[1, 51, 0.45, 1, 5] is a good choice.

Third-party Re-implementation

• Pytorch: FaceX-Zoo from JDAI.
• Pytorch: pt-femb-face-embeddings from Jonas Grebe

Logs

TODO list:

• add toy examples and release models
• migrate basic codes from the private codebase
• add beamer (after the ddl for iccv2021)
• test the basic codes
• add presentation
• migrate parallel training
• release mpu (Kaiyu Yue, in April) renamed to torchshard
• test parallel training
• add evaluation codes for recognition
• add evaluation codes for quality assessment
• add fp16
• test fp16
• extend the idea to CosFace, proved
• implement Mag-CosFace

20210909: add evaluation code for quality assessments

20210723: add evaluation code for recognition

20210610：[IMPORTANT] Mag-CosFace + ddp is implemented and tested!

20210601: Mag-CosFace is theoretically proved. Please check the updated arxiv paper.

20210531: add the 5-minutes presentation

20210513: add instructions for finetuning with MagFace

20210430: Fix bugs for parallel training.

20210427: [IMPORTANT] now parallel training is available (credits to Kaiyu Yue).

20210331 test fp32 + parallel training and release a model/log

20210325.2 add codes for parallel training as well as fp16 training (not tested).

20210325 the basic training codes are tested! Please find the trained model and logs from the table in Model Zoo.

20210323 add requirements and beamer presentation; add debug logs.

20210315 fix figure 2 and add gdrive link for checkpoint.

20210312 add the basic code (not tested yet).

20210312 add paper/poster/model and a toy example.

20210301 add ReadMe and license.