Skip to content
Multi-Task CNN for Pose-Invariant Face Recognition
C++ Python Cuda CMake MATLAB Makefile Shell
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
cmake
examples
imgs
include/caffe
matlab
python
scripts
src
tools
CMakeLists.txt
CONTRIBUTING.md
CONTRIBUTORS.md
INSTALL.md
LICENSE
Makefile
Makefile.config.example
README.md

README.md

MultiTask-CNN

Multi-Task CNN for Pose-Invariant Face Recognition.

Xi Yin and Xiaoming Liu.

Introduction

This repository contains the code to repeat the experiments on MultiPIE and CASIA-Webface as described in the paper. The major changes we have made in caffe is to split and merge batches based on the ground truth or estimated pose information.

We observe that for multi-task learning, it helps to learn joint features for all tasks. The final FC layer acts like feature selector to select the features for each specific task, thus it results in disentangled features, as illustrated in the following figure.

Multi-task Learning

Usage

After compile with caffe, please check the folder examples/ to see the prototxt files. We save data into HDF5 format.

Citation

If you found this code useful, please consider to cite:

@article{yin2018multi,
  title={Multi-task convolutional neural network for pose-invariant face recognition},
  author={Yin, Xi and Liu, Xiaoming},
  journal={IEEE Transactions on Image Processing},
  volume={27},
  number={2},
  pages={964--975},
  year={2018},
  publisher={IEEE}
}
You can’t perform that action at this time.