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

vicdu/CLM

CLM