PVANet-FACE: PVANET for face detection

Introduction

Training a face detection model using PVANet.

The dataset used for training is WIDERFACE

This repository contains source files of face detection using the PVANet. It is developed based on the awesome pva-faster-rcnn repository.

Requirement

1. Nivida CUDA 8.0
2. Nvidia CUDNN 6
3. Python 2

Installation

1. Clone this repository

   # Make sure to clone with --recursive
   git clone --recursive https://github.com/twmht/pva-faster-rcnn.git 

2. We'll call the directory that you cloned as FRCN_ROOT. Build the Cython modules

   cd $FRCN_ROOT/lib
   make

3. Build Caffe and pycaffe

   cd $FRCN_ROOT/caffe-fast-rcnn
   # Now follow the Caffe installation instructions here:
   #   http://caffe.berkeleyvision.org/installation.html
   # For your Makefile.config:
   #   Uncomment `WITH_PYTHON_LAYER := 1`
   
   cp Makefile.config.example Makefile.config
   make -j8 && make pycaffe

Training the face detection model

1. Download all available models (including pre-trained and compressed models)

   cd $FRCN_ROOT
   ./models/pvanet/download_all_models.sh

2. Download WIDERFace for training.

   I use python-widerface and cute-format to pack all the images of WIDERFace into the custom-defined imdb, where the format of imdb is different from VOC format.

   please look tools/convert_wider_to_imdb.py for detail.

   to run tools/convert_wider_to_imdb.py, update path to WIDERFace

   for example,

   # arg1: path to split (where the label file is)
   # arg2: path to images
   # arg3: path to label file name
    wider_train = WIDER('/opt/WiderFace/wider_face_split',
                  '/opt/WiderFace/WIDER_train/images',
                  'wider_face_train.mat')
   
    cw = CuteWriter('wider-imdb')
   
    run(wider_train, cw)

   this will generate a db named wider-imdb, and put wider-imdb into data/widerface/

3. Training PVANet

   cd $FRCN_ROOT
   tools/train_net.py --gpu 0 --solver models/pvanet/example_train/solver.prototxt --weights models/pvanet/pretrained/pva9.1_pretrained_no_fc6.caffemodel --iters 100000 --cfg models/pvanet/cfgs/train.yml --imdb wider

How to run the demo

1. Download pretrained model

2. Run the tools/demo.py

   cd $FRCN_ROOT
   ./tools/demo.py --net output/faster_rcnn_pvanet/wider/pvanet_frcnn_iter_100000.caffemodel --def models/pvanet/pva9.1/faster_rcnn_train_test_21cls.pt --cfg models/pvanet/cfgs/submit_1019.yml --gpu 0

Compression

If you want to compress your model, please look at tools/gen_merged_model.py. As compared to sanghoon's implementation (https://github.com/sanghoon/pva-faster-rcnn/blob/master/tools/gen_merged_model.py), I add the function to remove redundant power layers.