hysts/pytorch_mpiigaze
An unofficial PyTorch implementation of MPIIGaze and MPIIFaceGaze
What it solves
This project provides a PyTorch implementation of gaze estimation methods described in the MPIIGaze and MPIIFaceGaze papers. It allows researchers and developers to train and evaluate models that can determine where a person is looking based on image data.
How it works
The repository contains training and evaluation code that supports multiple neural network architectures, including LeNet, ResNet, and AlexNet. Users can download and preprocess the MPIIGaze and MPIIFaceGaze datasets, configure the model parameters via YAML files or command-line arguments using OmegaConf, and run training loops (such as leave-one-person-out evaluation) to measure the mean test angle error.
Who it’s for
This is primarily for researchers and developers working on computer vision and gaze estimation, specifically those looking to reproduce or build upon the original MPIIGaze and MPIIFaceGaze research papers.
Highlights
- Supports multiple model architectures: LeNet, ResNet-preact-8, AlexNet, and ResNet-14.
- Includes scripts for downloading and preprocessing the MPIIGaze and MPIIFaceGaze datasets.
- Uses OmegaConf for flexible configuration management.
- Provides a clear path to convert checkpoints to safetensors for use in a separate demo program (ptgaze).