EmotiW 2015 Challenge
Deep Learning for Emotion Recognition on Small Datasets Using Transfer Learning
Here we make available the code employed in our team’s submissions to the 2015 Emotion Recognition in the Wild contest, for the sub-challenge of Static Facial Expression Recognition in the Wild. The objective of this sub-challenge is to classify the emotions expressed by the primary human subject in static images extracted from movies.
We followed a transfer learning approach for deep Convolutional Neural Network (CNN) architectures. Starting from a network pre-trained on the generic ImageNet dataset, we performed supervised fine-tuning on the network in a two-stage process, first on datasets relevant to facial expressions, followed by the contest’s dataset.
Experimental results (see paper below) show that this cascading fine-tuning approach achieves better results, compared to a single stage fine-tuning with the combined datasets. Our best submission exhibited an overall accuracy of 48.5% in the validation set and 55.6% in the test set, which compares favourably to the respective 35.96% and 39.13% of the challenge baseline.
The code is released under a creative commons license. Our three best-performing models are also provided (Caffe platform is required in order to load and test them): submission3.zip (201MB), submission1.zip (201MB), submission8.zip (319MB). To reproduce the results from the contest, you will have to obtain a copy of the EmotiW dataset from the organizers.
The zip-files are encrypted; please email us for the password.
More details about this work can be found in the following paper:
Deep learning for emotion recognition on small datasets using transfer learning.
Proc. 17th ACM International Conference on Multimodal Interaction (ICMI), Emotion Recognition in the Wild Challenge, Seattle, WA, Nov. 9-13, 2015.
Please cite the above paper if you use our code.