This repository includes the code for training and evaluating the SSH face detector introduced in our ICCV 2017 paper.

The code is adapted based on an intial fork from the py-faster-rcnn repository.

If you find SSH useful in your research please consider citing:

@inproceedings{najibi2017ssh,
title={{SSH}: Single Stage Headless Face Detector},
author={Najibi, Mahyar and Samangouei, Pouya and Chellappa, Rama and Davis, Larry},
booktitle={The IEEE International Conference on Computer Vision (ICCV)},
year={2017}
}

1. Installation
2. Running the demo
3. Training a model
4. Evaluting a trained model

1. Clone the repository:

git clone --recursive https://github.com/mahyarnajibi/SSH.git

2. Install cuDNN and NCCL (used for multi-GPU training).

3. Caffe and pycaffe: You need to compile the caffe-ssh repository which is a Caffe fork compatible with SSH. Caffe should be built with cuDNN, NCCL, and python layer support (set by default in Makefile.config.example). You also need to make pycaffe.

4. Install python requirements:

pip install -r requirements.txt

5. Run make in the lib directory:

cd lib
make

To run the demo, first, you need to download the provided pre-trained SSH model. Running the following script downloads the SSH model into its default directory path:

bash scripts/download_ssh_model.sh