Multi-view Face Detection Using Deep Convolutional Neural Networks
Switch branches/tags
Nothing to show
Clone or download
Latest commit c139f26 Aug 20, 2016
Permalink
Failed to load latest commit information.
README.md Update README.md Aug 20, 2016
deploy.prototxt init version May 20, 2015
face_full_conv.prototxt add prob in rect May 21, 2015
face_rect.txt add face_rect.txt Jul 30, 2015
image_preprocess.py update May 28, 2015
solver.prototxt init version May 20, 2015
test.py update May 28, 2015
train.sh update May 28, 2015
train_val.prototxt update May 28, 2015

README.md

FaceDetection_CNN

Implement Yahoo Paper: Multi-view Face Detection Using Deep Convolutional Neural Networks

  1. Image Preprocess aflw dataset[1]. Use iou>=0.5 as positive, iou<=0.3 as negative. You should download the aflw dataset by yourself.

  2. Fine-tune Alex-Net using AFLW dataset. The model is in Baidu Yun: https://pan.baidu.com/s/1pJJ2WKN, or Google Drive: https://drive.google.com/file/d/0B8_dH3SiT7reMjJVRjJDXzJkRDQ/view?usp=sharing

  3. Convert fully connected layers into convolutional layers by reshaping layer parameters, see [2], you can use the convert_full_conv() function in test.py for converting.

  4. Get heat map for each scale of image.

  5. Process heat map by using non-maximal suppression to accurately localize the faces.

========== Reference: [1]https://lrs.icg.tugraz.at/research/aflw/

[2] http://nbviewer.ipython.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb