CAFFE implementation of my paper on low-resolution emotion recognition
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
experiments Delete .DS_Store Nov 10, 2017
pretrain Delete .DS_Store Nov 10, 2017
LICENSE Initial commit Oct 31, 2017

Robust Low-Quality Emotion Recognition

This is codes for Robust Emotion Recognition from Low Quality and Low Bit Rate Video: A Deep Learning Approach


  • compile CAFFE and matcaffe
  • install MATLAB


Download the data package for the Multimodal Affect Recognition Sub-Challenge (MASC) of the 6th Audio/Visual Emotion Challenge and Workshop (AVEC 2016): "Depression, Mood and Emotion".

  1. Use ./matlabscripts/prepare_dataset.m to generate cropped faces.

  2. Use ./matlabscripts/generate_h5/prepare_h5_files.m to generate HDF5 files for training

  3. Use network ./experiments/network/model_1F_CNN+D.prototxt and solver ./experiments/solver/solver_1F_CNN+D.prototxt for HQ and LQ

  4. Use ./pretrain/generate_train.m and ./pretrain/generate_test.m to generate HDF5 files for SR model

  5. Use ./pretrain/Pretrain_solver.prototxt to train SR model

  6. Use ./experiments/network/model_1F_CNN+D_vlqr_non_joint.prototxt and ./experiments/solver/solver_1F_CNN+D_vlqr_non_joint.prototxt for LQ-non-joint

  7. Use ./experiments/network/model_1F_CNN+D_vlqr.prototxt and ./experiments/solver/solver_1F_CNN+D_vlqr.prototxt for vlqr

Pretrain Model

You can download the pretrained model from Dropbox


If you use this code for research, please cite our papers:

  title={Robust emotion recognition from low quality and low bit rate video: A deep learning approach},
  author={Cheng, Bowen and Wang, Zhangyang and Zhang, Zhaobin and Li, Zhu and Liu, Ding and Yang, Jianchao and Huang, Shuai and Huang, Thomas S},
  journal={arXiv preprint arXiv:1709.03126},
  title={Enhance Visual Recognition under Adverse Conditions via Deep Networks},
  author={Liu, Ding and Cheng, Bowen and Wang, Zhangyang and Zhang, Haichao and Huang, Thomas S},
  journal={arXiv preprint arXiv:1712.07732},