OmniCompass: 820-Video Benchmark Evaluates 50 MLLMs on Omni-Modal Video Understanding
_akhaliq · x · 2026-10-17
PolyU VCLab released OmniCompass (HF Paper of the Day), a holistic benchmark for omni-modal video grounding and understanding: 820 videos, 2,460 instances, 1,800 human annotation hours. It spans OmniCompass-VL and OmniCompass-AV tracks, three tasks (temporal grounding, spatial grounding, QA), and paired text vs text+image target specifications. Evaluating 50 MLLMs shows open-source models still lag proprietary ones (widest in audio-visual), with precise spatiotemporal grounding and fine-grained perception remaining key weaknesses.
Related event: PolyU Releases OmniCompass Benchmark for Omnimodal Video Grounding(2 posts)→
More from Multimodal
- Real-time video labeling demo tracks a basketball shot's full trajectory live — LinusEkenstam · 2026-10-17
- Open-source image model MOSS-Image flies under radar, hailed as 'the new SDXL' — flowersslop · 2026-10-17
- Plenio 1.0: free open-source AI song studio for ComfyUI ships with a player that renders videos — Vivid_Promise1700 · 2026-10-17
- AI-rendered 'Tokyo Night' Roadster wallpapers fool viewers into thinking they're real photos — JOBhakdi · 2026-10-17
- ComfyUI-SceneCut: frame-exact scene detection, cut fixing and clip export for ComfyUI — LatentSpacer · 2026-10-17
- AI video workflow: reroll at 480p then one $1 upscale to 4K cuts costs 87% — FellMentKE · 2026-10-17