1. GitHub

https://github.com/playerkk/face-py-faster-rcnn

2. Requirements

Requirements: software

Requirements for caffe and pycaffe (see: Caffe installation instructions)

Python packages you might not have: cython, python-opencv, easydict 

Requirements: hardware

For training smaller networks (ZF, VGG_CNN_M_1024) a good GPU (e.g., Titan, K20, K40, ...) with at least 3G of memory suffices
For training Fast R-CNN with VGG16, you'll need a K40 (~11G of memory)
For training the end-to-end version of Faster R-CNN with VGG16, 3G of GPU memory is sufficient (using CUDNN)

3. Installation

1. Clone the face Faster R-CNN repository
   git clone –recursive https://github.com/playerkk/face-py-faster-rcnn.git

2. Build the Cython modules
   cd $FRCN_ROOT/lib
   make

3. Build Caffe and pycaffe
   cd $FRCN_ROOT/caffe-fast-rcnn
   make -j8 && make pycaffe

4. 下载预先训练好的VGG模型
   A pre-trained face detection model trained on the WIDER training set is available here.
   http://supermoe.cs.umass.edu/%7Ehzjiang/data/vgg16_faster_rcnn_iter_80000.caffemodel
   放置目录：
   $FRCN_ROOT/output/faster_rcnn_end2end/train/vgg16_faster_rcnn_iter_80000.caffemodel

5. 下载测试数据
   http://vis-www.cs.umass.edu/fddb/index.html 下载FDDB数据库放入$FRCN_ROOT/data目录：
   包括:
   FDDB
   FDDB/FDDB-folds
   FDDB/originalPics

4.Test the trained model

python ./tools/run_face_detection_on_fddb.py –gpu=0
运行完成后显示：

十组图片，每检测完11张图片显示完成度 XX%

在run_face_detection_on_fddb.py 添加保存图片的命令

# for j in xrange(dets.shape[0]): 下面添加以下代码
 p1 = (int(dets[j, 0]), int(dets[j, 1]))
 p2 = (int(dets[j, 0] + dets[j, 2]), int(dets[j, 1] + dets[j, 3]))
 cv2.rectangle(im, p1, p2, (0, 0, 255))
 cv2.imwrite("/home/dl/faceBox.jpg", im)

效果：

---

---

---

---

---

---

---

---

---

5. 自己训练模型

1. Download pre-computed Faster R-CNN detectors
   cd $FRCN_ROOT
   ./data/scripts/fetch_faster_rcnn_models.sh

2. Download the WIDER face dataset. Extract all files into one directory named WIDER
   http://mmlab.ie.cuhk.edu.hk/projects/WIDERFace/
   WIDER/
   WIDER/WIDER_train/
   WIDER/WIDER_val/

3. Download the (http://jianghz.me/files/wider_face_train_annot.txt) and put it under the WIDER directory.

4. Create symlinks for the WIDER dataset
   cd FRCNROOT/dataln−s WIDER WIDER

5. Follow the next sections to download pre-trained ImageNet models
   cd $FRCN_ROOT
   ./data/scripts/fetch_imagenet_models.sh

6. To train a Faster R-CNN face detector using the approximate joint training method, use experiments/scripts/faster_rcnn_end2end.sh. Output is written underneath $FRCN_ROOT/output.

   cd FRCN_ROOT
   cd FRCN_ROOT
   ./experiments/scripts/faster_rcnn_end2end.sh [GPU_ID] [NET] wider [–set …]

   eg:
   ./experiments/scripts/faster_rcnn_end2end.sh 0 VGG16 wider

7. Trained Fast R-CNN networks are saved under: （GTX980训练了10多个小时）
   output/ experiment directory / dataset name /
   

6. 遇到的问题

http://blog.csdn.net/zengdong_1991/article/details/51614315