Generalizing Gaze Estimation with Weak-Supervision from Synthetic Views

The implementation of Arxiv paper for gaze estimation task.

Preparation

1. Download the dataset and put it under data/

2. Download eyes3d.pkl and put it under assets/

3. Download pretrained checkpoint and put it under assets/

4. Install libraries:

   Copy

   pip install timm pytorch-lightning==1.8.1 albumentations==1.3.0
   

Testing with pre-trained model

After downloading the pre-trained checkpoint above,

python test_gaze.py assets/latest_a.ckpt

Training

python trainer_gaze.py

Results