Face-Transformer

This is the code of Face Transformer for Recognition (https://arxiv.org/abs/2103.14803v2).

Recently there has been great interests of Transformer not only in NLP but also in computer vision. We wonder if transformer can be used in face recognition and whether it is better than CNNs. Therefore, we investigate the performance of Transformer models in face recognition. The models are trained on a large scale face recognition database MS-Celeb-1M and evaluated on several mainstream benchmarks, including LFW, SLLFW, CALFW, CPLFW, TALFW, CFP-FP, AGEDB and IJB-C databases. We demonstrate that Transformer models achieve comparable performance as CNN with similar number of parameters and MACs.

Usage Instructions

1. Preparation

The code is mainly adopted from Vision Transformer, and DeiT. In addition to PyTorch and torchvision, install vit_pytorch by Phil Wang, and package timm==0.3.2 by Ross Wightman. Sincerely appreciate for their contributions. All needed Packages are found in requirements.txt -> Simply install all packages by:

pip install -r requirements.txt

Copy the files of folder "copy-to-vit_pytorch-path" to vit_pytorch path.

.
├── __init__.py
├── vit_face.py
└── vits_face.py

2. Databases

You can download the training databases, MS-Celeb-1M (version ms1m-retinaface), and put it in folder 'Data'.

You can download the testing databases as follows and put them in folder 'eval'.

LFW: Baidu Netdisk(password: dfj0), Google Drive
SLLFW: Baidu Netdisk(password: l1z6), Google Drive
CALFW: Baidu Netdisk(password: vvqe), Google Drive
CPLFW: Baidu Netdisk(password: jyp9), Google Drive
TALFW: Baidu Netdisk(password: izrg), Google Drive
CFP_FP: Baidu Netdisk(password: 4fem), Google Drive--refer to Insightface
AGEDB: Baidu Netdisk(password: rlqf), Google Drive--refer to Insightface

3. Train Models

ViT-P8S8

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VIT -head CosFace --outdir ./results/ViT-P8S8_ms1m_cosface_s1 --warmup-epochs 1 --lr 3e-4 

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VIT -head CosFace --outdir ./results/ViT-P8S8_ms1m_cosface_s2 --warmup-epochs 0 --lr 1e-4 -r path_to_model 

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VIT -head CosFace --outdir ./results/ViT-P8S8_ms1m_cosface_s3 --warmup-epochs 0 --lr 5e-5 -r path_to_model

ViT-P12S8

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VITs -head CosFace --outdir ./results/ViT-P12S8_ms1m_cosface_s1 --warmup-epochs 1 --lr 3e-4 

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VITs -head CosFace --outdir ./results/ViT-P12S8_ms1m_cosface_s2 --warmup-epochs 0 --lr 1e-4 -r path_to_model 

CUDA_VISIBLE_DEVICES='0,1,2,3' python3 -u train.py -b 480 -w 0,1,2,3 -d retina -n VITs -head CosFace --outdir ./results/ViT-P12S8_ms1m_cosface_s3 --warmup-epochs 0 --lr 5e-5 -r path_to_model

4. Pretrained Models and Test Models (on LFW, SLLFW, CALFW, CPLFW, TALFW, CFP_FP, AGEDB)

You can download the following models

ViT-P8S8: Baidu Netdisk(password: spkf), Google Drive
ViT-P12S8: Baidu Netdisk(password: 7caa), Google Drive

You can test Models

The content of “property” file for “ms1m_retinaface” dataset is as follows: "93431,112,112"