Kornia's DoG/Hessian now match OpenCV quality, with Astra improving results in one shot
ducha_aiki · x · 2026-09-06
Kornia maintainer duchaaiki reports that the open-source vision library's DoG and Hessian feature detectors now match OpenCV/vlfeat in quality at comparable speed, and a follow-up notes that Astra improved the results further in one shot.
He also candidly noted the earlier work was neither a one-shot Claude job nor fully manual — it took a week of back-and-forth, an honest data point on the real difficulty of AI-assisted open-source contributions.
More from coding & agent
- Agent takes over a Twitter thread and proactively cleans up slack along the way — andimarafioti · 2026-09-06
- Astra's video editing: auto-cuts bad takes, alt-tabs, and dead footage — adonis_singh · 2026-09-06
- Astra plays Redstone and edits its own gameplay videos — adonis_singh · 2026-09-06
- Benchmarking agent workflows: frontier vs cheaper models trade-offs — zainhas · 2026-09-06
- Two devs fix 136 bugs and add 4 company integrations in just 3 days — cneuralnetwork · 2026-09-06
- Developer lets coding agent run overnight and wakes up to a fixed bug — TianbaoX · 2026-09-06