Bikefitting

AI-powered bike fitting analysis. Detects bike angle and rider joint angles from video.

What's in each folder

models/ - All the AI models live here. You need these to run anything.

create_labeled_dataset/ - Tool to create training data by syncing video with phone IMU data.

bike_angle_detection_model/ - Train and run the bike angle prediction model.

joint_angle_detection/ - Live webcam joint angle detection.

bikefitting-web/ - Web app for video analysis (upload a video, get results).

sam3/ - External library (already installed, don't touch).

Setup

You need Python 3.10+ and the models folder with all 3 model files.

Windows

python -m venv bikeEnv
bikeEnv\Scripts\activate
pip install --upgrade pip
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
pip install -e sam3

Mac/Linux

conda create -n bikefitting python=3.12
conda activate bikefitting
./install.sh

Quick start

1. Make sure the models/ folder has all 3 files (best_model.pt, yolov8m-pose.pt, yolov8n-seg.pt)

2. Try live joint detection:

   cd joint_angle_detection
   python 1_live_camera.py
   

3. Or run the web app - see bikefitting-web/README.md