VISTA (Video Grounding) - Referral Queries: leaderboard

VISTA (2026) spatio-temporal video grounding with referral queries, which name only the subject and its attributes. 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 referral queries (the subject and its attributes only, extracted from the caption by an LLM), 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 8B62.85#917
2VISTA CogVLM Grounding (checkpoint unspecified)60.56
3InternVL2.5-8B51.11#2370
4VISTA Sphinx-v2 (checkpoint unspecified)47.79
5MiniGPT-v246.62
6Qwen-VL-Chat45.56#7860
7VISTA MiMo-VL-7B (checkpoint unspecified)43.34
8Shikra-7B30.91
9LLaVA-Grounding-7B22.51
10Ferret-7B17.74

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