VSI-Bench

From the catalogue. Listed by one researcher from one source page.

• id: vsi-bench
• kind: language and reasoning annotation
• robot or device: egocentric videos of real indoor 3D scan datasets (class: sensor only)
• how collected: QA derived from 3D scene annotations, human reviewed
• size as stated: over 5,000 QA pairs from 288 real videos
• hours: not stated | hours only claimed: none | episodes: not stated
• year: 2024
• organisation: nyu-visionx
• licence: apache-2.0 (class: open)
• access: open (class: open)
• link (dataset): https://huggingface.co/datasets/nyu-visionx/VSI-Bench
• Hugging Face repository: nyu-visionx/VSI-Bench, 7,967 downloads in 30 days, https://huggingface.co/datasets/nyu-visionx/VSI-Bench
• note: Standard visual-spatial intelligence benchmark for embodied VLMs.

Confirm the licence at the link before relying on it.