What drawing lines on football pitch taught me about future of human-AI collab
2 days ago
- #AI-human collaboration
- #computer vision
- #sports analysis
- The author initially tried to use Codex to analyze a soccer goal, but the AI's initial results were inaccurate despite appearing convincing.
- Human domain expertise was crucial; the author realized known football dimensions could calibrate measurements, which Codex hadn't suggested.
- Incorporating human-in-the-loop feedback improved accuracy, as manually marking points in a workbench reduced errors from automated detection.
- Codex introduced concepts like homography for handling perspective distortion in player movement analysis, enabling better collaboration between human intuition and AI engineering.
- Leveraging context-specific tools (like Apple Vision APIs on Mac) instead of default Python methods produced superior results, such as cleaner player cutouts for visualizations.