Software
UpstreamDrift (featured — this site's primary software)
UpstreamDrift (formerly Golf Modeling Suite) is the unified golf-swing modeling
platform this website documents. It provides a FastAPI + React web UI, a REST API,
and a suite of end-user tools (ball flight, motion capture / C3D import, character
builder, putting/bunker/terrain, video analyzer, RL training) across multiple
physics engines (Pinocchio, MuJoCo, Drake; OpenSim/MyoSuite experimental).
Launch it with python launch_upstream_drift.py.
Follow the proximal-to-distal worked program from the governed technical monograph, through the models, claims, and evidence at protected UpstreamDrift commit 85cce4d3 (opens in a new tab), to the interactive workbench. This is a computational research example; independent review and participant-level human validation remain open.
OpenSim
OpenSim is a freely available, open-source software platform for modeling, simulating, and analyzing the neuromusculoskeletal system. Developed at Stanford University, OpenSim enables researchers to build and analyze musculoskeletal models, perform inverse kinematics and dynamics, and simulate movement. It is widely used in biomechanics research, rehabilitation, and sports science applications, including golf swing analysis.
Visit OpenSim → (opens in a new tab)MyoSim
Open-source software for simulating muscle-driven movement, providing tools for biomechanical modeling and analysis.
View MyoSim → (opens in a new tab)OpenPose
OpenPose is a real-time multi-person pose estimation library developed by Carnegie Mellon University. It can detect human body, hand, facial, and foot keypoints on single images or video streams. OpenPose is valuable for biomechanics research, motion capture, and analyzing human movement patterns, making it useful for golf swing analysis and sports biomechanics applications.
View OpenPose → (opens in a new tab)
Drake
Drake is an open-source toolbox developed by the Robot Locomotion Group at MIT. It provides a suite of tools for analyzing the dynamics of robots and building control systems. Drake is designed to provide a flexible and efficient platform for robotics research and development, with capabilities for modeling, simulation, trajectory optimization, and control design. Highly relevant for implementing control-affine systems and analyzing underactuated systems.
Visit Drake → (opens in a new tab)MuJoCo
MuJoCo (Multi-Joint dynamics with Contact) is a physics engine for model-based optimization, robotics, biomechanics, graphics and animation, and other areas where fast and accurate simulation is needed. MuJoCo provides state-of-the-art contact dynamics, efficient computation, and a comprehensive API for building and simulating complex multibody systems.
Visit MuJoCo → (opens in a new tab) MuJoCo Playground → (opens in a new tab)Pinocchio
Pinocchio is a fast and flexible implementation of rigid body dynamics algorithms and their analytical derivatives. Developed by the Stack of Tasks team, Pinocchio provides efficient algorithms for forward/inverse dynamics, Jacobian computations, and constraint handling for multibody systems. It is particularly useful for robotics applications requiring fast dynamics computations and is based on the algorithms described in Roy Featherstone's "Rigid Body Dynamics Algorithms."
Visit Pinocchio → (opens in a new tab)Machine Learning Software
GolfPose
GolfPose is a deep learning framework for golf swing pose estimation, presented at ICPR 2024. The system provides 2D and 3D pose estimation for golfers and golf clubs, with models for detection, 2D keypoint estimation, and 3D pose reconstruction. It includes pre-trained models for golfer detection, club detection, and pose estimation using architectures like HRNet, ViTPose, and DEKR. The framework supports both individual golfer/club pose estimation and combined golfer-with-club pose estimation, making it a valuable tool for golf biomechanics analysis and swing motion capture.
View on GitHub → (opens in a new tab) Read Paper → (opens in a new tab)SwingNet
SwingNet is a lightweight deep neural network with a hybrid deep convolutional and recurrent architecture designed for golf swing sequencing. Introduced alongside the GolfDB dataset, SwingNet detects eight key events in golf swings at an average rate of 76.1%, with six out of eight events detected at 91.8%. The model is computationally efficient, making it suitable for deployment on mobile devices for in-the-field analysis. SwingNet enables automated event detection in golf swings, facilitating consistent evaluation and analysis of swing mechanics.
Read Paper → (opens in a new tab)Golf Swing Analysis (GitHub Search)
A curated search for open-source repositories related to golf swing analysis using computer vision. Results vary over time; individual repositories should be evaluated for code quality and maintenance status before use.
Search on GitHub → (opens in a new tab)Golf Swing Pose Estimation (GitHub Search)
A curated search for open-source pose estimation tools applied to golf swing analysis. These repositories typically use computer vision frameworks to extract joint angles and body segment positions from video. Results vary over time.
Search on GitHub → (opens in a new tab)