:loudspeaker: This is cloned repository!

This repository is cloned from Tencent/FaceDetection-DSFD and modified for research, compatibility, functionality and convenience.

Update

• 2019.04: Release pytorch-version DSFD inference code.
• 2019.03: DSFD is accepted by CVPR2019.
• 2018.10: Our DSFD ranks No.1 on WIDER FACE and FDDB

Introduction

In this repo, we propose a novel face detection network, named DSFD, with superior performance over the state-of-the-art face detectors. You can use the code to evaluate our DSFD for face detection.

For more details, please refer to our paper DSFD: Dual Shot Face Detector! or poster slide!

Our DSFD face detector achieves state-of-the-art performance on WIDER FACE and FDDB benchmark.

WIDER FACE

Easy, Medium and Hard denote AP on WIDER FACE validation set Easy, Medium and Hard, respectively.

E2E latency denotes an end-to-end latency (= preprocess + network + TTA + postprocess).

Latency is measured with batch size 1 on RTX 2080Ti GPU and Threadripper 2950X CPU.

Confidence thresholds were set to 0.01 for both AP and latency benchmark.

FDDB

Requirements

• cudatoolkit==10.2
• cudnn==7.6
• python==3.6
• torch==1.4.0
• torchvision==0.5.0

Development Environments

Conda

• conda-py36torch14.yml is appropriate if you'd like to use this repository on conda environment
• Please refer to Managing environments — conda documentation for more details

Docker

• nvcr.io/nvidia/pytorch:19.11-py3 is appropriate if you'd like to use this repository on docker container
• Please refer to PyTorch | NVIDIA NGC for more details

Getting Started

Installation

• Clone this repository

> git clone https://github.com/swoook/dsfd.git
> cd ${DSFD_DIR}/dsfd

Demo

• Refer to demo.md for detailed instructions

WIDER FACE Validation

1. Data Preparation : Refer to data-preparation.md for detailed instructions
2. Evaluation: Refer to validation.md for detailed instructions

Qualitative Results

Citation

If you find DSFD useful in your research, please consider citing:

@inproceedings{li2018dsfd,
  title={DSFD: Dual Shot Face Detector},
  author={Li, Jian and Wang, Yabiao and Wang, Changan and Tai, Ying and Qian, Jianjun and Yang, Jian and Wang, Chengjie and Li, Jilin and Huang, Feiyue},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year={2019}
}

Contact

For any question, please file an issue or contact

Jian Li: swordli@tencent.com