The paper introduces MIRA-Bench, a mirror‑centric RGB‑D dataset and benchmark designed to test vision‑language models (VLMs) in reflective scenes where depth information is misleading. It contains 1,011 RGB‑D images from 525 real‑world scenes, annotated with reflective surface masks and object labels for both direct and reflected views, and 14,140 QA pairs across eight sub‑tasks. Evaluation shows that state‑of‑the‑art VLMs consistently fail to distinguish real from reflected objects, misalign reflected instances with their real counterparts, and produce incorrect spatial inferences, with depth data even degrading performance. whyItMatters":"The study reveals that current VLMs cannot handle mirror‑induced depth ambiguity, highlighting a critical limitation in spatial reasoning for real‑world applications involving reflective surfaces."
arXiv:2607. 00491v1 Announce Type: cross Abstract: Benchmarks for vision-language models (VLMs) mostly test observational spatial reasoning: models describe relations already visible in the input.
arXiv:2601. 19099v2 Announce Type: replace-cross Abstract: Vision--language models (VLMs) achieve strong performance on many multimodal benchmarks but remain brittle on spatial reasoning tasks that require aligning abstract overhead representations with egocentric views.
SpaceCast-Bench is a new benchmark that evaluates predictive spatial reasoning in vision‑language models, moving beyond simple spatial perception to tasks that require constructing scenes, anticipating interventions, and reasoning about unseen outcomes. It contains 3,862 questions from 182 real‑world scenes across 16 task types and three difficulty levels—static perception, local prediction, and global prediction—testing scene understanding, spatial state updating, and relational inference. Evaluation of 21 models shows a large performance gap, with the best model achieving only 58.0% versus 87.2% human accuracy, and fine‑tuning on the benchmark’s programmatically generated data can substantially improve model performance.
arXiv:2609.16233v1 Announce Type: cross Abstract: Vision-language models excel at 2D image understanding but remain limited in 3D spatial reasoning. Progress is hindered by limitations in current ben...
arXiv:2605. 20448v2 Announce Type: replace-cross Abstract: Vision-language models reliably name objects in a scene, but do they represent the 3D layout those objects inhabit?
arXiv:2605.30557v2 Announce Type: replace-cross Abstract: Spatial reasoning benchmarks typically evaluate whether vision-language models can derive the correct answer from a visual observation. Yet i...
As Multimodal Large Language Models (MLLMs) are increasingly deployed in decision-critical pipelines such as robotics, embodied AI, and safety monitoring, the opacity of their spatial judgments limits operator trust and auditability. MLLMs demonstrate strong reasoning but often struggle with fine-grained spatial understanding and object hallucination.
Vision-language models excel at 2D image understanding but remain limited in 3D spatial reasoning. Progress is hindered by limitations in current benchmarks. First, 3D datasets often rely on point clo...
Metric-Bench introduces a new benchmark for Vision‑Language Models (VLMs) that focuses on metric‑spatial reasoning in indoor scenes by using in‑image reference objects with known dimensions. The accompanying MetricReasoner fine‑tuning recipe employs structured prompts and numerical rewards to implicitly learn 2D‑to‑3D mapping without camera intrinsics. Experiments show that this approach improves spatial metric understanding by 43.1 % over existing models and boosts downstream embodied tasks, while also delivering gains on general VLM benchmarks.
arXiv:2610.23416v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) have significantly advanced instruction-driven perception through the "Embedding-as-Mask" paradigm. However, e...
The paper introduces PCSR-Bench, a benchmark of 84,373 question‑answer pairs derived from 2,600 omnidirectional images across 26 indoor environments, designed to evaluate perspective‑conditioned spatial reasoning (PCSR) in multimodal large language models (MLLMs). It reports a significant perception–reasoning gap, with accuracy dropping from 57.59% on limited field‑of‑view reasoning to as low as 0.64% on open‑ended compositional directional chains. An RL‑based diagnostic study on a 7B‑scale model shows that reward shaping can improve performance to 60.06% on a controlled task, indicating partial plasticity of PCSR capabilities.
The paper introduces Geometry-Privileged Distillation (GPD), a method that incorporates depth, semantic, and bird’s-eye-view cues as compact text routed to a teacher model during on-policy self-distillation. By applying a privileged KL loss only to incorrect trajectories, GPD enhances spatial reasoning in vision‑language models while keeping the deployed model RGB‑only. Experiments on a 4B backbone show that GPD outperforms prior methods on multiple spatial reasoning benchmarks, with ablations confirming the benefits of 3D cues, question‑conditioned routing, and selective distillation.