Michael Black on academic CV research's role in the age of powerful large models
CSProfKGD · x · 2026-09-10
- Michael Black (attending ECCV since 1992) reflects on the role of academic computer vision research as large models grow ever more powerful, and how researchers can still have impact.
- His example: VIGA, which takes an image and outputs a 3D Blender scene — a classic inverse-graphics task and the first solved with an agentic approach. The idea is years old; an early rejection significantly delayed publication before this year's ECCV acceptance.
- Yutaka Asano (FunAILab) adds that labs also struggle with papers that 'work' but aren't 'novel', and argues vision researchers should ask whether the next model will solve their task anyway, and design methods that could actually enable MLLMs to get there.
Related event: Michael Black asks: what's left for academic computer vision in the LLM era(5 posts)→
More from AGI Musings
- Ex-DeepMind researcher: many colleagues truly believe AI could kill everyone — Dan_Jeffries1 · 2026-09-10
- iamtrask: alignment is a data problem, and humanity is '100% gonna survive and thrive' — iamtrask · 2026-09-10
- Hank Yang: AI Excels at Well-Defined Problems, So the Real Skill Is Defining New Ones — hankyang94 · 2026-09-10
- dlwh: every eval is broken, yet blindly hillclimbing on them drove stunning AI progress — dlwh · 2026-09-10
- AI safety debate: if you believe your P(doom), why aren't you fighting? — MichaelAlleged · 2026-09-10
- Blogger Claims OpenAI Could Lure Scientists With Free Compute to Mine Their Research Ideas — ns123abc · 2026-09-10