Streetlifting Dip Validator

⚠️ WORK IN PROGRESS: This is an early prototype. Features may change and accuracy is not guaranteed.

AI-powered tool to analyze parallel bar Dips and determine lift validity based on depth rules.

---

📸 Examples

✅ VALID Dip

D (Green) touches/passes below E (Blue) line VALID (+10.0px margin)

❌ INVALID Dip

D fails to reach E line INVALID (−2.3px margin)

---

What It Does

Analyzes smartphone video of parallel bar Dips and outputs:

• VALID / INVALID decision based on depth rule
• Annotated overlay video with landmarks and phases
• JSON report with margins, confidence, and per-frame trace

The Rule

A dip is VALID if the posterior deltoid (D) reaches or goes below the elbow line (E) at any point.

margin = y_D - y_E  (Y increases downwards)
VALID = max(margin) >= 0

---

How It Works

Step	Description
Pose Estimation	RTMPose (17 keypoints) via rtmlib
Landmark Refinement	Geometric estimation of D and E from arm vectors
Phase Detection	Savitzky-Golay smoothing on depth signal
Decision	Best margin across entire video

---

Usage

# Install
pip install -e .

# Run
python -m dip_validator input_videos/video.mp4

Output

output/<video_name>/
├── overlay.mp4          # Annotated video
├── report.json          # Full analysis data
├── debug_landmarks.jpg  # Bottom frame visualization
└── debug_pose.jpg       # Pose keypoints

---

Tech Stack

• Python 3.10+
• rtmlib (RTMPose)
• OpenCV, NumPy, SciPy