Practical overview
When it fits
Use it when you have a padel match recording from a fixed camera and want locally generated tracking and performance metrics, with the ability to configure model checkpoints and mark court keypoints before inference.
Who it is for
- Padel coaches and players reviewing movement and stroke patterns from match footage.
- Broadcast teams seeking match analytics for graphics or audience-facing information.
- Developers or analysts who want match-derived data for later database storage.
Documented uses
- Track player and ball positions and estimate their velocities from a recording.
- Review player pose estimates and classifications, including examples such as forehand and backhand volleys.
- Generate heatmaps and two-dimensional court projections for match review.
- Inspect stroke-associated ball velocity and player error rates.
- Derive analytics that may inform broadcast graphics or later data storage.
Requirements & limits
The project is presented as an open-source public GitHub repository with clone, setup and inference instructions. The documented application is self-hosted and command-line driven.
- The documented workflow assumes a fixed camera setup and requires manual selection of 12 court keypoints.
- Model weights must be downloaded and their checkpoint paths set in config.py.
- Default batch sizes require at least 8 GB of VRAM; the README says batch sizes can be reduced in config.py.
- Setup uses Python 3.12, Conda, the repository requirements and PyTorch.
- The README describes possible broadcast and database uses, but does not document built-in integrations for either.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the local inference environment
Prepare the repository, dependencies and model checkpoint configuration for running inference.
Before you start
- A Python 3.12 Conda environment
- PyTorch installed according to the PyTorch installation instructions
- Model weights and paths to the checkpoints
Steps
Clone the repository.
Create and activate the Conda environment, then install the requirements with pip.
Install PyTorch and download the model weights.
Edit config.py to specify the model checkpoint paths.
Expected outcome
The environment and model paths are configured for the repository’s documented inference stage.
Guide limitations
- The setup steps do not by themselves run inference.
- The documented inference command is python main.py from the repository root.
- Default batch sizes require at least 8 GB of VRAM; the configuration can be adjusted.
- Inference assumes a fixed camera setup and requires selection of 12 court keypoints.
Run inference on a padel recording
Start the repository’s documented inference workflow after setup and configuration.
Before you start
- The repository and dependencies set up
- Model weights downloaded and checkpoint paths configured in config.py
- A fixed-camera padel recording
- At least 8 GB of VRAM for default batch sizes
Steps
From the repository root, run python main.py.
When prompted, select 12 unique court keypoints for the homographic computations.
Expected outcome
The application runs its computer-vision inference workflow to derive match analytics from the recording.
Guide limitations
- The README does not specify a detailed output-file workflow.
- The camera setup is assumed to be fixed.
- Default batch sizes require at least 8 GB of VRAM; reduce them in config.py if needed.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- Joao-M-Silva/padel_analytics · Official homepage title · retained in its source language