OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding

arXiv:2610.18829 · cs.CV, cs.CL, cs.MM · Submitted 2026-10-13 · Read on arXiv

cs.CV, cs.CL, cs.MM

Submitted: 2026-10-13

Updated: 2026-10-13

Terminology

• Benchmark — 1× in this paper · explained in 2 episodes
• Grounding — 1× in this paper · explained in 3 episodes

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models