Awesome

<div align="center"> <h2> DVGaze: Dual-view Gaze Estimation <a href='https://arxiv.org/abs/2308.10310.pdf'><img src='https://img.shields.io/badge/ArXiv-PDF-red' style="vertical-align:middle;"></a> </h2> <div> <i><a href="https://yihua.zone/">Yihua Cheng </a> and <a href='https://scholar.google.com.hk/citations?user=9ggbm0QAAAAJ&hl=en'>Feng Lu</a>, ICCV 2023</i> </div> <br> <img src="images/teaser.png" width = "400" height = "200" alt="图片名称" align=center /> </div>

Description

Dual cameras have been applied in many devices recently. In this paper, we explore a new direction for gaze estimation. We propose a dual-view gaze estimation network (DV-Gaze) including dual-view interactive convolution block and dual-view transformers. DVGaze

Usage

Please re-write the config file in config/train/xxx.yaml and config/test/xxx.yaml.

To train a model, Please run the command:

python trainer/total.py config/train/xxx.yaml

To evaluate a model, please run the command:

python tester/total.py config/train/xxx.yaml config/test/xxx.yaml

where config/train/xxx.yaml indicates the config for training model.

Please download the label file from <a href='https://drive.google.com/drive/folders/16yt3xjkQzR_hA5EMFWQhrL-s2f3A3MKb?usp=sharing'> Google Driver </a>. We are not authorized to distribute image files. Please access <a href='https://phi-ai.buaa.edu.cn/Gazehub/'> Gazehub </a> to acquire data-processing code on ETH and EVE. Note that the website only works on working time (8:00-19:00?) in the China time zone (I have graduated and don't know the reason). Apologise for the inconvenience. I will try my best to update data-processing codes in Github.

All codes are beta version and we will keep updating this repositories. The full version will be updated before Oct. 10.

Please send email to y.cheng.2@bham.ac.uk if you have any questions.

We can also meet in ICCV conference in person:) Have a good day.