https://github.com/lebellig/emotions-gcn
https://github.com/lebellig/emotions-gcn
Last synced: about 1 month ago
JSON representation
- Host: GitHub
- URL: https://github.com/lebellig/emotions-gcn
- Owner: lebellig
- Created: 2022-03-09T20:40:49.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2022-03-09T20:56:05.000Z (over 4 years ago)
- Last Synced: 2025-09-08T12:47:52.342Z (11 months ago)
- Language: Jupyter Notebook
- Size: 3.24 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# emotions-gcn
Implementing Graph Convolutional Networks (GCNs) for emotion classification. Original graphs are made using the Delaunay method to convert face-landmarks into triangular mesh. Prior study showed that only some landmarks are relevant for emotion classification so we do not select the others.
Moreover, we propose a preprocessing pipeline to make our model robust to face size variations and implemented data-augmentation (rotation, flip, face stretching, +gaussian noise).
All neural networks blocks and models can be found in the `src/models` folder whereas preprocessing and datasets are located in `src/data`.
The model reach 0.657 accuracy on the validation set.