RealityBench: Simulated Environments to Evaluate a Vision Language Model's Ability to Document Human Physical Work
cs.CV, cs.AI, cs.CL
Submitted: 2026-10-09
Updated: 2026-10-14
Sources
- Are Large Pre-trained Vision Language Models Effective Construction Safety Inspectors?
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning
- An Egocentric Vision-Language Model based Portable Real-time Smart Assistant
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