# gaze-point-estimation-2023

**Repository Path**: elorfiniel/gaze-point-estimation-2023

## Basic Information

- **Project Name**: gaze-point-estimation-2023
- **Description**: Codebase for Gaze Point Estimation using GazeCapture-like Datasets.
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2024-02-14
- **Last Updated**: 2026-04-28

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Gaze Point Estimation 2023

Codebase for Gaze Point Estimation using GazeCapture-like Datasets.

```text
gaze_point_estimation_2023/
 ├─ checkpoints/      # the trained models
 ├─ datasets/         # dataset configs
 ├─ source/
 │   ├─ litmodels/    # pytorch lightning modules
 │   ├─ models/       # pytorch modules
 │   ├─ run_spec/     # train and test configs
 │   ├─ utils/        # dataset, dataloader, etc.
 │   └─ <scripts>.py  # main scripts
 └─ requirements.txt  # dependencies
```

## Preprocessing

Run the preprocessing script for GazeCapture dataset, using the following command:

```shell
python preprocess_gaze_capture.py \
  --dataset_path /path/to/unzipped_gaze_capture \
  --output_path /path/to/output/directory \
  --use_parallel --n_workers 16 \
  --use_recovery --use_validation \
  --face_resize "(224,224)" \
  --eyes_resize "(224,224)" \
```

The preprocessing script will create a new directory with the following structure:

```text
path/to/output/directory/
 ├─ mean_face.npy           # mean image for all face crops
 ├─ mean_reye.npy           # mean image for all reye crops
 ├─ mean_leye.npy           # mean image for all leye crops
 └─ <recording>
     ├─ mean_face_crop.npy  # mean face crop for this recording
     ├─ mean_reye_crop.npy  # mean reye crop for this recording
     ├─ mean_leye_crop.npy  # mean leye crop for this recording
     ├─ metadata_info.txt   # dataset split and device info
     ├─ metadata_labels.npy # gaze annotations
     ├─ metadata_thetas.npy # face alignment angles
     ├─ face_crop/*.jpg     # face crops
     ├─ reye_crop/*.jpg     # reye crops
     ├─ leye_crop/*.jpg     # leye crops
     └─ landmark/*.npy      # landmark files
```

## Dataset Configs

See `datasets/dataset_spec` for detailed description on dataset preparation.

## Runner Configs

See `source/run_spec` for detailed description on training and testing configs.

## Running

Please use the following steps to train or test the model:

1. Configure the dataset, eg. `datasets/dataset_spec/dataset_example_a.toml`.
2. Configure the runner, eg. `source/run_spec/run_example_a.toml`.
3. Adjust the run script, eg. `source/example_run.py`.
4. Start running, eg. `python source/example_run.py --run_spec <run_spec>`

## Notes

Please note that the code provided in this repository represents only a portion of the complete project. Due to the terms of our cooperation agreement, some critical parts of the codebase have not been uploaded. This repository serves to showcase certain components and functionalities developed under the scope of the project.