What Looks Like a Capability Limit in Vision-Language Models Is a Readout Limit
cs.CV, cs.AI, cs.CL
Submitted: 2026-09-23
Updated: 2026-09-23
Terminology
Sources
- Qwen2.5-VL Technical Report
- Improving LLM First-Token Predictions in Multiple-Choice Question Answering via Output Prefilling
- PointArena: Probing Multimodal Grounding Through Language-Guided Pointing
- Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision?
- Qwen3-VL Technical Report
- Do Vision-Language Models See or Guess? Measuring and Reducing Textual-Prior Reliance with a Phrasing-Controlled Benchmark
- LocateAnything: Fast and High-Quality Vision-Language Grounding with Parallel Box Decoding
- Improving GUI Grounding with Explicit Position-to-Coordinate Mapping
- Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text
- GUI-Actor: Coordinate-Free Visual Grounding for GUI Agents
- MMGist: A Comprehensive Multimodal Benchmark for 2027
- IoU-PD: IoU-Aware Privileged Distillation for Visual Grounding with Multimodal Large Language Models
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