Hill Climb Gesture-Based Control

Simple OpenCV-based Hill Climb Racing game control using MediaPipe Hand Landmarker.

Quantized TFLite Pipeline

This project now uses the MediaPipe Tasks HandLandmarker pipeline backed by a TFLite model bundle (hand_landmarker.task).

• Default mode: float16.
• Int8 path is temporarily disabled in code while conversion is stabilized.

Model Selection

Use environment variables:

• HAND_LANDMARKER_VARIANT=float16 (default: float16)
• HAND_LANDMARKER_MODEL_PATH=... to use any local .task model file
• HAND_LANDMARKER_MODEL_URL=... optional download URL when model file is missing

Notes:

• Public MediaPipe hand landmarker bundle is currently available as float16.
• Int8 references are intentionally disabled for now.

Run

Install dependencies:

pip install mediapipe opencv-python

Start the controller:

python main.py

Run (auto-download float16 from MediaPipe):

set HAND_LANDMARKER_VARIANT=float16
python main.py

Startup Self-Check Output

At startup, the app prints model verification details, including:

• requested variant
• model source
• resolved model path
• model size
• model last-modified timestamp (UTC)
• SHA-256 hash

This helps confirm that the project is actually running the expected model.

Press q to quit.