/opal23_headpose

Headpose estimation using OPAL (2023)

Primary LanguageJupyter NotebookMIT LicenseMIT

On the representation and methodology for wide and short range head pose estimation

PWC PWC PWC

We provide Python code in order to replicate the head pose experiments in our paper https://doi.org/10.1016/j.patcog.2024.110263

If you use this code for your own research, you must reference our journal paper:

On the representation and methodology for wide and short range head pose estimation
Alejandro Cobo, Roberto Valle, José M. Buenaposada, Luis Baumela.
Pattern Recognition, PR 2024.
https://doi.org/10.1016/j.patcog.2024.110263

Requisites

Installation

This repository must be located inside the following directory:

images_framework
    └── alignment
        └── opal23_headpose

Usage

usage: opal23_headpose_test.py [-h] [--input-data INPUT_DATA] [--show-viewer] [--save-image]

  • Use the --input-data option to set an image, directory, camera or video file as input.

  • Use the --show-viewer option to show results visually.

  • Use the --save-image option to save the processed images.

usage: Alignment --database DATABASE
  • Use the --database option to select the database model.
usage: Opal23Headpose [--gpu GPU]
  • Use the --gpu option to set the GPU identifier (negative value indicates CPU mode).
usage: Opal23Headpose [--rotation-mode {euler,quaternion,6d,6d_opal}]
  • Use the --rotation-mode option to specify the internal pose parameterization of the network.
> python images_framework/alignment/opal23_headpose/test/opal23_headpose_test.py --input-data images_framework/alignment/opal23_headpose/test/example.tif --database 300wlp --gpu 0 --rotation-mode euler --save-image

Notebooks

The directory notebooks contains examples regarding some of the contributions of our paper:

  • Loading annotations: this notebook contains a simple implementation of a PyTorch Dataset class that uses our annotations for CMU Panoptic Dataset and shows sample images.

  • Prediction alignment: this notebook shows an example on how to align model predictions to reduce systematic errors in cross-dataset evaluations.

  • Opal loss: this notebook shows an usage example of our Opal Loss function for Head Pose Estimation.