Official Implementation

Figure: Illustration of performance comparison in terms of gaze error and tunable parameters. Our method achieved the lowest gaze error with negligible tunable parameters required for personalization.

• Python 3.8.0
• Pytorch 1.10.2
• CUDA 10.2

ETH-XGaze, Gaze360, MPIIFaceGaze, and EyeDiap datasets are downloaded from their respective official websites.

The datasets are pre-processed using the method outlined in [1]. The code for data pre-processing is released by the authors of [1] and is available at this page.

reader.py consists of the dataloader class, please note the dataloader works with the specific format of the pre-processed datasets. The dataloader class needs to be modified accordingly if using another format.

python3 main.py --train_status=train --train_loader_flag=XXX --model_save_dir="XXX" --data_dir="XXX" --resnet_model_path="XXX"

python3 main.py --train_status=test --val_loader_flag=XXX --test_data_dir="XXX" --resnet_model_path="XXX" --pre_trained_model_path="XXX"

python3 main.py --train_status=meta_train --train_loader_flag=XXX --val_loader_flag=XXX --batch_size=20 --model_save_dir="XXX" --data_dir="XXX" --test_data_dir="XXX" --resnet_model_path="XXX" --pre_trained_model_path="XXX" --pre_trained_model_name="XXX"

python3 main.py --train_status=persona --val_loader_flag=XXX --test_data_dir="XXX" --resnet_model_path="XXX" --pre_trained_model_path="XXX" --pre_trained_model_name="XXX"

Example usages are provided in run_gaze360.sh and run_xgaze.sh, please modify the directories and paths accordingly.

If you find this paper or code is helpful in your work, please consider citing:

@inproceedings{liu2024test, title={Test-Time Personalization with Meta Prompt for Gaze Estimation}, author={Liu, Huan and Qi, Julia and Li, Zhenhao and Hassanpour, Mohammad and Wang, Yang and Plataniotis, Konstantinos N and Yu, Yuanhao}, booktitle={Proceedings of the AAAI Conference on Artificial Intelligence}, volume={38}, number={4}, pages={3621--3629}, year={2024} }

[1] Cheng, Y.; Wang, H.; Bao, Y.; and Lu, F. 2021. Appearance-based Gaze Estimation With Deep Learning: A Review and Benchmark. arXiv preprint arXiv:2104.12668.