task_categories:
- other
tags:
- computer-vision
- gaze-estimation
- virtual-reality
VRGaze: A Large-scale Dataset for VR Gaze Estimation
VRGaze is the first large-scale off-axis gaze estimation dataset for Virtual Reality (VR), introduced in the paper "GazeShift: Unsupervised Gaze Estimation and Dataset for VR".
Dataset Summary
The dataset comprises 2.1 million near-eye infrared images collected from 68 participants. It is specifically designed to address data scarcity in VR gaze research, focusing on the off-axis camera configurations typical of modern headsets.
- Images: 2.1 million infrared images.
- Participants: 68 individuals.
- Hardware: Off-axis camera configurations common in modern VR systems.
- Purpose: Designed for unsupervised gaze representation learning and few-shot calibration.
Dataset Sample
VRGaze dataset sample:
Usage
To use this dataset with the official GazeShift implementation, follow these steps:
Installation
git clone https://github.com/gazeshift3/gazeshift
cd gazeshift
pip install -r requirements.txt
Training
To reproduce the experiments on the VRGaze dataset, run the provided training script (ensure you update the dataset and output locations within the script):
bash Train.sh
The model is expected to achieve a mean angular error of approximately 1.84° after per-person calibration (around 400K steps).