CLM

"Constrained Local Models" based on "Mastering OpenCV Chapter6".

I refactored all sources and add warping demo. Tracking accuracy is not so high, but more readable code than original to study CLM.

Requirements

• Linux (testing Ubuntu 14.04 and Arch Linux)
• OpenCV 3.1 (2.4 is also supported if replaced link options in premake5.lua)
• premake 5

Build

Edit premake5.lua for your environment.

premake5 gmake
cd build
make

Following commands can be executed in build directory.

Run with trained model

./bin/release/main

To configure cascade path, set new path with --cascade argument, and CLM model also can be changed with --clm (default model is trained using helen dataset). Web camera will be used by default, and image can be used with --image.

Train

Now, MUCT dataset is available to train (helen mode is broken). Please download MUCT dataset, expand it and execute following command (Change paths).

./bin/release/train --out OUTPUT/DIR --muct_image_dir YOUR/muct/jpg/ --muct_lm_file YOUR/muct/muct-landmarks/muct76-opencv.csv

Visualized shape

Visualized patch

Visualized detector

Warping demo 1

yukiti is a demo program which tracks your face using web camera and warps Japanese bill.

./bin/release/yukiti

Warping demo 2

yukiti2 is also a demo program. This replaces your face with bill's one.

./bin/release/yukiti2