OpenCap-Offline

A fully offline, localized implementation of Stanford NMBL's OpenCap. This version adapts the original cloud-based processing pipeline to run entirely on a local machine, ensuring strict data privacy and allowing for use in environments without internet access. This version of the software requires a sufficiently powerful GPU, but cannot run on blackwell architecture cards (i.e. Nvidia 50-series cards). I recommend use with Nvidia 30-series or 40-series cards with minimum 12GB graphics ram.

🔔 Stay Updated

Never miss a new feature or bug fix. Click the Watch button at the top right of this GitHub page, select Custom, and check the box for Releases. GitHub will automatically send you an email notification whenever a new version of OpenCap Offline is published!

Prerequisites

• Python 3.9.25
• Git
• NVIDIA GPU (RTX 3000 series or 4000 series recommended) - this setup is not yet suitable for GPUs with Blackwell architecture
• Anaconda or Miniconda

Installation & Setup

Due to a number of critical version dependencies, please ensure you complete the following install steps in order:

1. Download the Code
Clone this repository to your local machine:
git clone https://github.com/driscollh/opencap-offline.git

2. Create a Python Environment
conda create -n opencap_env python=3.9 -y
conda activate opencap_env

3. Install the Correct Chumpy Version
Navigate into the main folder:
pip install chumpy==0.70 --no-build-isolation --no-deps

4. Install CPython and OpenSim Packages
Navigate into the main folder and install the required Python environment packages, followed by opensim:
conda install -c opensim-org opensim=4.4 -y
pip install --no-cache-dir -r requirements.txt

5. Download the Local Dependencies (Required)
Because the machine learning models and background engines (OpenPose, FFmpeg) are too large for GitHub, they are hosted securely on Zenodo.

• Go to the V3 Zenodo archive: https://doi.org/10.5281/zenodo.19447679
• Download the dependencies.zip file.
• Extract the contents directly into the empty dependencies/ folder inside this project. This should result in three separate subfolders, one for each key dependency.
• Extract the RTMPose.zip file directly into the main project folder next to pyqt5_launcher_improved.py. This folder should be titled "Blackwell_RTMPose", and contain a dependencies/ folder and mmcv/ folder.

Usage

Once the environment is set up and the dependencies are in place, you can launch the local processing by running the main Python script in your environment:

python pyqt5_launcher_improved.py or python simple_launcher.py

VideoRecording

Unlike the online version of OpenCap, this offline version requires additional video capture for calibration of camera intrinsics and extrinsics.
Please consult the recording_practices.pdf file prior to data collection.

Acknowledgments and Licensing

This project is built upon the original open-source OpenCap project by the Stanford Neuromuscular Biomechanics Laboratory.

• Original OpenCap: https://github.com/opencap-org/opencap-core
• License: Distributed under the Apache 2.0 License.

Citation

If you use OpenCap Offline in your research or clinical workflow, please cite it using the following DOI:

APA:

Driscoll, H. G. (2026). OpenCap Offline (Version 2.2.2) [Computer software]. https://doi.org/10.5281/zenodo.20621570

BibTeX:

@software{opencap_offline_2026,
  author       = {Driscoll, Harry G.},
  title        = {OpenCap Offline},
  month        = jun,
  year         = 2026,
  publisher    = {Zenodo},
  version      = {2.2.2},
  doi          = {10.5281/zenodo.20621570},
  url          = {[https://doi.org/10.5281/zenodo.20621570](https://doi.org/10.5281/zenodo.20621570)}
}