Code for Emotion Recognition in the Wild (EmotiW) challenge

For EmotiW 2017 code please see folder 2017

cd to folder 2017, then run the code as: python3 submission5.py

The code will download all necessary features precomputed using our models, train emotion classification models and predict emotions on the test set.

Test accuracy is 60.03% (second place).

• Python3 (necessary to load features saved in python3)

• Ubuntu 14.04 LTS

Challenge website

Paper: Convolutional neural networks pretrained on large face recognition datasets for emotion classification from video accepted to the 1st Workshop on Large-scale Emotion Recognition and Analysis. IEEE International Conference on Automatic Face and Gesture Recognition 2018, Xi'an, China

If you use our code or features in your work, please cite our paper as following:

@article{knyazev2017convolutional,
  title={Convolutional neural networks pretrained on large face recognition datasets for emotion classification from video},
  author={Knyazev, Boris and Shvetsov, Roman and Efremova, Natalia and Kuharenko, Artem},
  journal={arXiv preprint arXiv:1711.04598},
  year={2017}
}