Discarded or Unreachable? Diagnosing and Repairing Spatial Relation Failure in Vision--Language Similarity
cs.CV, cs.AI
Submitted: 2026-11-03
Updated: 2026-11-03
Code: https://github.com/mahdims/vl-spatial-moment
Terminology
Sources
- LEACE: Perfect linear concept erasure in closed form
- Understanding Hardness of Vision-Language Compositionality from A Token-level Causal Lens
- Amnesic Probing: Behavioral Explanation with Amnesic Counterfactuals
- On the Binding Problem in Artificial Neural Networks
- What's left can't be right -- The remaining positional incompetence of contrastive vision-language models
- CLIP Behaves like a Bag-of-Words Model Cross-modally but not Uni-modally
- Scaling Open-Vocabulary Object Detection
- Frame Averaging for Invariant and Equivariant Network Design
- How and where does CLIP process negation?
- Qwen2.5 Technical Report
- SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features
- A Good CREPE needs more than just Sugar: Investigating Biases in Compositional Vision-Language Benchmarks
- A Theory of Usable Information Under Computational Constraints
- Left-Right Symmetry Breaking in CLIP-style Vision-Language Models Trained on Synthetic Spatial-Relation Data
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