How abhijitshukla/spinvision works
Project Type and Purpose
• This is a data-processing pipeline (computer-vision) application written in Python. Its goal is to take an input cricket video, detect the ball’s frame-by-frame position, simulate a no-spin trajectory from the bounce point, and overlay actual vs. predicted paths to estimate spin deviation.
File Structure Analysis
Top-level scripts and folders reveal a clear, linear pipeline:
• coord.py → Detection step using a YOLOv8 model
• predict.py → Physics-based trajectory simulation (no spin)
• overlay.py → Visualization & video output
• model/ → Contains the pre-trained YOLOv8 detector (.pt)
• videos/ → Raw input videos
• coordinates/ → Intermediate coordinate files (actual & predicted)
• output/ → Final overlay videos & snapshots
• results/ → Example outputs for reference
Key Components
a. Input Source
• “videos/” directory (e.g., kohli.mp4)
b. Detection Module
• coord.py
• YOLOv8 model (model/cricket_ball_detector.pt)
• Output: coordinates/coordinates.txt
c. Bounce & Physics Module
• predict.py
• Reads coordinates.txt, computes post-bounce trajectory without spin
• Output: coordinates/coordinates_no_spin.txt
d. Visualization Module
• overlay.py
• Reads both coordinate files + input video
• Renders green (actual) vs. red (predicted) trajectories
• Outputs overlay_output.mp4 & snapshot_with_spin_angle.png
e. Data Stores
• coordinates/
• output/
• results/ (for reference)
f. External Dependencies
• Python runtime, OpenCV, YOLOv8 framework, NumPy, etc.
Relationships and Data Flow
• Feedback loops for parameter tuning or reruns
Architectural Patterns & Principles
• Pipeline/Batch-processing pattern
• Separation of concerns: detection, simulation, visualization in distinct scripts
• Intermediate, filesystem-based data passing (coordinate text files)
• Reproducibility: clear inputs/outputs for each stage