[WIP]Face Alignment at 3000 FPS via Regressing Local Binary Features

Work in Progress

• [x] Training local binary features
• [x] Training global linear regression
• [x] Model serialization
• [x] Validation

Referrences

• Face Alignment at 3000 FPS via Regressing Local Binary Features
• One Millisecond Face Alignment with an Ensemble of Regression Trees
• Face Alignment by Explicit Shape Regression
• Robust Face Landmark Estimation under Occlusion
• freesouls/face-alignment-at-3000fps

Requirements

• Python 3
• OpenCV 3
• Boost 1.64+
• Dlib
• OpenMP

Installation

macOS

Python 3

brew install python3

Boost

brew install boost-python --with-python3

OpenCV 3

brew tap homebrew/science
brew install opencv3 --with-python3　--without-python

Open /usr/local/Cellar/opencv/3.3.1/lib/pkgconfig/opencv.pc and replace -llibopencv_xxxxxx.3.3.1.dylib with -lopencv_xxxxxx.

Ubuntu

Boost

./bootstrap.sh --with-python=python3 --with-python-version=3.5
./b2 python=3.5 -d2 -j4 --prefix YOUR_BOOST_DIR install

Dlib

sudo apt install libomp-dev
pip install dlib --user

Datasets

IBUG

Download afw.zip, ibug.zip, lfpw.zip, helen.zip from ibug.doc.ic.ac.uk/resources/facial-point-annotations/

or from Google Drive.

• afw.zip
• ibug.zip
• lfpw.zip
• helen.zip

┣━━ afw
┃      ┣━━ 134212_1.jpg
┃      ┣━━ 134212_1.pts
┃             ...
┣━━ helen
┃      ┣━━ trainset
┃      ┃      ┣━━ 232194_1.jpg
┃      ┃      ┣━━ 232194_1.pts
┃      ┃             ...
┃      ┗━━ testset
┃              ┣━━ 232194_1.jpg
┃              ┣━━ 232194_1.pts
┃                     ...
┣━━ ibug
┃      ┣━━ image_003_1.jpg
┃      ┣━━ image_003_1.pts
┃             ...
┗━━ lfpw
        ┣━━ trainset
        ┃      ┣━━ image_0001.jpg
        ┃      ┣━━ image_0001.pts
        ┃             ...
        ┗━━ testset
                ┣━━ image_0001.jpg
                ┣━━ image_0001.pts
                       ...