Ahmed A.Abdelrahman, Thorsten Hempel, Aly Khalifa, Ayoub Al-Hamadi, ICIP, 2022, "under review". [Arxiv]
Human gaze is a crucial cue used in various applications such as human-robot interaction and virtual reality. Recently, convolution neural network (CNN) approaches have made notable progress in predicting gaze direction. However, estimating gaze in-the-wild is still a challenging problem due to the uniqueness of eye appearance, lightning conditions, and the diversity of head pose and gaze directions. In this paper, we propose a robust CNN-based model for predicting gaze in unconstrained settings. We propose to regress each gaze angle separately to improve the per-angel prediction accuracy, which will enhance the overall gaze performance. In addition, we use two identical losses, one for each angle, to improve network learning and increase its generalization. We evaluate our model with two popular datasets collected with unconstrained settings. Our proposed model achieves state-of-the-art accuracy of 3.92 and 10.41 on MPIIGaze and Gaze360 datasets, respectively.
If you use any part of our code or data, please cite our paper.
@misc{AAbdelrahman2022L2CSNetFG,
title={L2CS-Net: Fine-Grained Gaze Estimation in Unconstrained Environments},
author={Ahmed A.Abdelrahman, Thorsten Hempel, Aly Khalifa, Ayoub Al-Hamadi},
year={2022},
eprint={2203.03339},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
- Set up a virtual environment:
python3 -m venv venv
source venv/bin/activate
- Install required packages:
pip install -r requirements.txt
- Install the face detector:
pip install git+https://github.com/elliottzheng/face-detection.git@master
- Download the pre-trained models from here and Store it to models/.
- Run:
python demo.py \
--snapshot models/L2CSNet_gaze360.pkl \
--gpu 0 \
--cam 0 \
This means the demo will run using L2CSNet_gaze360.pkl pretrained model
We provide the code for train and test MPIIGaze dataset with leave-one-person-out evaluation.
- Download MPIIFaceGaze dataset from here.
- Apply data preprocessing from here.
- Store the dataset to datasets/MPIIFaceGaze.
python train.py \
--dataset mpiigaze \
--snapshot output/snapshots \
--gpu 0 \
--num_epochs 50 \
--batch_size 16 \
--lr 0.00001 \
--alpha 1 \
This means the code will perform leave-one-person-out training automatically and store the models to output/snapshots.
python test.py \
--dataset mpiigaze \
--snapshot output/snapshots/snapshot_folder \
--evalpath evaluation/L2CS-mpiigaze \
--gpu 0 \
This means the code will perform leave-one-person-out testing automatically and store the results to evaluation/L2CS-mpiigaze.
To get the average leave-one-person-out accuracy use:
python leave_one_out_eval.py \
--evalpath evaluation/L2CS-mpiigaze \
--respath evaluation/L2CS-mpiigaze \
This means the code will take the evaluation path and outputs the leave-one-out gaze accuracy to the evaluation/L2CS-mpiigaze.
We provide the code for train and test Gaze360 dataset with train-val-test evaluation.
-
Download Gaze360 dataset from here.
-
Apply data preprocessing from here.
-
Store the dataset to datasets/Gaze360.
python train.py \
--dataset gaze360 \
--snapshot output/snapshots \
--gpu 0 \
--num_epochs 50 \
--batch_size 16 \
--lr 0.00001 \
--alpha 1 \
This means the code will perform training and store the models to output/snapshots.
python test.py \
--dataset gaze360 \
--snapshot output/snapshots/snapshot_folder \
--evalpath evaluation/L2CS-gaze360 \
--gpu 0 \
This means the code will perform testing on snapshot_folder and store the results to evaluation/L2CS-gaze360.