lesson_50_synthetic_hand_pose_estimation

Lesson 50: Synthetic Hand Pose Estimation

This toy lesson renders a compact grayscale hand skeleton and regresses ten normalized keypoints: one wrist anchor, a thumb tip, and four finger base-tip pairs. The renderer is deterministic given the seed and stays light enough for CPU-only smoke tests.

The training loop follows the same regression shell used by neighboring vision lessons and writes the standard artifacts expected by the repo's lesson runner.

Run:

python -m tracks.vision.lesson_50_synthetic_hand_pose_estimation.train --epochs 1 --device cpu

Outputs land under outputs/vision/lesson_50_synthetic_hand_pose_estimation/<run_name>/.