Contributors:
- Bhagesh Gaur (2020558)
- Madhava Krishna (2020217)
- Sanyam Goyal (2020116)
The project involves multilabel classification using the ODIR dataset.
The detailed report can be found here: DL Report Final.
The broad steps involved are:
- Preprocessing the data to extract single eye-features (credits to Jordi Corbella).
- Creating histogram equalised images of the eye-images.
- Train Swin-T transformer with early-fusion of image and other features (age, sex).
- Using an ensemble of Swin-T vision transformers.