Simple CurricularFace

This repository is a simplified version of the official PyTorch implementation of CurricularFace.
This will be useful for those who want to train the model on a custom dataset or with the specific option.

• Kang Dong Won & Won Bin

• Kim Suk Jin & Min Yun Ki

Usage

test

1. download checkpoint from here
2. simple test

python scripts/test.py

3. make a grid

python scripts/grid.py

Dataset

1. Download MS-Celeb-1M-Align_112x112 (85K ids/5.8M imgs) from here
2. unzip the zip file and make a lst file.

python im2rec.py --list --resize 112 --recursive {prefix} {path_to_imgs} 
ex) python im2rec.py --list --resize 112 --recursive ./my_data ../dataset/ms1m_align_112_imgs

train

1. set path in "config.py"

- line 5: DATA_ROOT = '~/dataset/ms1m_align_112_imgs', # the parent root where your data are stored
- line 6: RECORD_DIR = '~/CurricularFace/my_data.lst', # the dataset record dir

2. train

python scripts/train.py

• Note that we modified some lines in the original code as below.

  • use wandb instead of tensorboard
  • add a line, "if cfg['RANK'] % ngpus_per_node == 0:", when a multi processing error occurs
  • fix a few lines in 'dataset.py' as below:

- line 71: image_dir = line[0]--> image_dir = line[2]
- line 74: image_dir, label = line[0], line[1] --> image_dir, label = line[2], line[1]
- line 75: label = int(label) --> label = int(float(label))

Contributors to this repository

Yukyeong Lee | yukyeongleee@gmail.com
Wonjong Ryu | 1zong2@innerverz.com

Reference

CurricularFace

Citing CurricularFace

If you find this code useful in your research, please consider citing the authors:

@article{huang2020curricularface,
	title={CurricularFace: Adaptive Curriculum Learning Loss for Deep Face Recognition},
	author={Yuge Huang and Yuhan Wang and Ying Tai and  Xiaoming Liu and Pengcheng Shen and Shaoxin Li and Jilin Li, Feiyue Huang},
	booktitle={CVPR},
	pages={1--8},
	year={2020}
}