Study finds no visual foundation model fully maintains spatial state

Researchers from NYU, NTU, HKUST and CUHK probed 10 visual foundation models across four levels of cross-observation spatial state maintenance, finding dramatic performance gaps, with none capable of fully supporting spatial state.

2026-11-02 ~ 2026-11-02 · 3 related posts

• Probing 10 vision foundation models finds none supports a full spatial state — liuziwei7 · 2026-11-02
• Probing shows visual foundation models vary widely in persistent spatial state support — mmlab-ntu · 2026-11-02
• New probe tests 10 vision foundation models on persistent spatial state across 4 levels — liuziwei7 · 2026-11-02