VISTA (Video Grounding) - Human-Self: leaderboard
VISTA (2026) spatio-temporal video grounding on queries describing solitary human actions. 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 solitary human actions (human-self), 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B | 75.7 | #917 |
| 2 | Qwen-VL-Chat | 58.8 | #7860 |
| 3 | InternVL2.5-8B | 50.9 | #2370 |
| 4 | VISTA Sphinx-v2 (checkpoint unspecified) | 48.1 | |
| 5 | VISTA CogVLM Grounding (checkpoint unspecified) | 46.8 | |
| 6 | MiniGPT-v2 | 46.4 | |
| 7 | VISTA MiMo-VL-7B (checkpoint unspecified) | 46 | |
| 8 | LLaVA-Grounding-7B | 38.9 | |
| 9 | Shikra-7B | 38.8 | |
| 10 | Ferret-7B | 33.6 |
Interactive version: theaggregate.ai/benchmark?slug=vista-video-grounding-human-self · How It Works · Data refreshed daily, snapshot 2026-11-21.