Where Do Vision-Language Models Get Left and Right? Position-Code Counterfactuals in Qwen2.5-VL, Qwen3-VL and InternVL3
cs.CV, cs.LG
Submitted: 2026-11-10
Updated: 2026-11-10
Code: https://github.com/chen4study-hub/where-vlms-get-lef
Terminology
Sources
- Decodable Is Not Grounded: A Vision-Ablation Arbiter for VLM Spatial Reasoning
- Why Do MLLMs Struggle with Spatial Understanding? A Systematic Analysis from Data to Architecture
- The Dual Mechanisms of Spatial Variable Binding in Vision-Language Models
- Beyond Semantics: Rediscovering Spatial Awareness in Vision-Language Models
- Visual Credit Audit for Multimodal Spatial Reasoning
- Seeing Depth but Choosing Cues in Vision-Language Models
- VitaGlyph: Vitalizing Artistic Typography with Flexible Dual-branch Diffusion Models
- OrientSAM: Mitigating Camera-Centric Shortcut in Multimodal Spatial Reasoning via Orientation-Aware Spatial Alignment
- Qwen2.5-VL Technical Report
- How Vision Transformers Represent Position: A Mechanistic Comparison of Absolute and Rotary Positional Encodings
- Geometry without Position? When Positional Embeddings Help and Hurt Spatial Reasoning
- Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution
- Left-Right Symmetry Breaking in CLIP-style Vision-Language Models Trained on Synthetic Spatial-Relation Data
- How to use and interpret activation patching
- Pathways of Visual Information Flow in Vision-Language Models
- Who Is Left of Whom? Tracing Spatial Evidence and Role Binding in Relative-Position Reasoning
- Readout Blindness: VLM Scores Miss the Spatial Direction Their Frozen Encoders Retain
Related papers
- TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models
- Do Egocentric Video-Language Models Capture Both Hand- and Object-Centric Cues?
- Aura: Consistent Multi-Subject Video Generation via VLM-Grounded Semantic Alignment
- PVRF: All-in-one Adverse Weather Removal via Prior-modulated and Velocity-constrained Rectified Flow
- DLWM: Diverse Latent World Models for Efficient Multimodal Reasoning
- Out of Sight, Still in Mind: Token Compression for Omni-LLMs