NeurIPS paper finds vision models fail to recognize objects without their typical neighbours
FrancescoLocat8 · x · 2026-10-04
A NeurIPS 2026 paper, "Stranger Things: When Objects Appear Without Their Typical Neighbours", examines typicality bias across modern vision models: can a model still recognize an object when its usual contextual neighbours disappear? The thread suggests models depend heavily on typical co-occurring contexts, degrading on out-of-context objects.
More from Research
- Yoav Goldberg: RL tasks like winning games or replicating bibliographies teach models facts — yoavgo · 2026-10-04
- Frontier VLMs Struggle at Basic 2-DoF Active Visual Search, NeurIPS Paper Finds — _vztu · 2026-10-04
- Cognit scores LLMs on famous trick questions like the bat-and-ball problem, with a human-mode twist — UnemployedTechie2021 · 2026-10-04
- NYU's $5 open-source eFlesh magnetic tactile sensor is 3D-printable — lukas_m_ziegler · 2026-10-04
- Lelarge: the AI maths paper cited even the first prior work — marc_lelarge · 2026-10-04
- SPEAR Simulator Opens Up 14K+ Unreal Engine Functions for Embodied AI Research — rsasaki0109 · 2026-10-04