Skip to content

StoryMY/RefractionGaze

Repository files navigation

RefractionGaze

The repository of paper "Towards Eyeglasses Refraction in Appearance-based Gaze Estimation". (ISMAR 2023)

Requirements

The code is tested with Python 3.7 and torch 1.8.1.

To install the other packages:

pip install -r requirements.txt

Data

This repository contains data of the Relative Coordinate Representation (RCR) and eye region images. Please download the data from Google Drive and put them in the dataset directory.

The data is organized as follows:

dataset
|-- 001
    |-- 001_000_01.h5
    |-- 001_000_02.h5
    |-- 001_550_01.h5
|-- 002
    |-- ...
    ...
|-- 006
    |-- ...

Train on non-glass data

# take ID-001 for example
python main.py --config configs/001.json

Test on testing data

# take ID-001 for example
# dim: 0 for pitch, 1 for yaw
# type: 0 for naked-eye results, 1 for corrected results
python draw_test_results.py --data <path_to_npy> --dim 1 --type 1

About

Official pytorch implementation of paper "Towards Eyeglasses Refraction in Appearance-based Gaze Estimation" (ISMAR 2023).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages