VISTA (Video Grounding) - Spatial: leaderboard

VISTA (2026) spatio-temporal video grounding on samples whose primary demand is spatial, positional configurations among entities. Mean spatio-temporal IoU in percent; higher is better.

Metric: Mean spatio-temporal IoU (m_vIoU, %): per-frame box IoU summed over the frames where predicted and true tubes overlap in time, divided by their temporal union, on the samples the taxonomy labels spatial (positional configurations among entities), over the 11,814 video-query pairs of VISTA (HCSTVG-v1 and v2, VidVRD, VidSTG, MeViS and RVOS videos re-annotated with an interaction taxonomy), zero-shot on sub-sampled frames; the model must localize the queried subject with a box in every frame; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 10 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 VL 8B64.8#917
2VISTA CogVLM Grounding (checkpoint unspecified)57.5
3InternVL2.5-8B46.3#2370
4Qwen-VL-Chat45.7#7860
5MiniGPT-v243.1
6VISTA Sphinx-v2 (checkpoint unspecified)42.6
7VISTA MiMo-VL-7B (checkpoint unspecified)36.9
8Shikra-7B29.9
9LLaVA-Grounding-7B28.1
10Ferret-7B20.9

Interactive version: theaggregate.ai/benchmark?slug=vista-video-grounding-spatial · How It Works · Data refreshed daily, snapshot 2026-11-21.