arXiv:2608. 14138v1 Announce Type: cross Abstract: Spatial perception and reasoning from visual observations require recovering geometric structure, establishing correspondences, and understanding spatial relations.
arXiv:2606. 11918v1 Announce Type: new Abstract: Current Large Reasoning Models (LRMs) exhibit remarkable general capabilities but significantly underperform in spatial reasoning tasks.
VisualSphinx is a large-scale synthetic dataset designed to train vision language models (VLMs) on visual logical reasoning tasks. The dataset is generated through a rule-to-image synthesis pipeline that expands puzzle rules from seed questions and produces executable rendering programs for puzzle assembly. Experiments using standard GRPO with binary correctness rewards show that VLMs trained on VisualSphinx achieve improved logical coherence and performance on logical reasoning, as well as on related tasks such as algebraic, arithmetic, and geometry reasoning.
MV-STRIDE is a Multi‑View hierarchical Spatial Reasoning dataset that models dependencies among perception, scene understanding, and contextual reasoning to support 3D spatial cognition. It introduces a QA generation pipeline that enforces cross‑view constraints, producing multi‑level reasoning tasks with chain‑of‑thought supervision. Experiments show that training on MV‑STRIDE yields state‑of‑the‑art performance on multi‑view spatial benchmarks, enabling MLLMs to reason robustly across diverse viewpoints.
The paper introduces a progressive training strategy for embodied vision‑language models aimed at reducing spatio‑temporal hallucinations. It first creates a Chain‑of‑Thought dataset that breaks complex reasoning into detailed spatiotemporal steps, then uses supervised pre‑training on this dataset followed by fine‑tuning with weakly‑labeled data. Experiments show the method improves backbone accuracy and narrows the forward‑backward performance gap from over 70% to 6.53%, indicating stronger dynamic reasoning and fewer temporal biases.
The paper introduces ReImaGin, a method that uses image generation models as a flexible visual reasoning tool for multimodal large language models. Unlike traditional fixed-function vision tools, ReImaGin accepts natural language commands and can perform open-ended visual operations such as removing occlusions or creating floorplans from multiple views. Experiments on six diverse visual reasoning tasks show that ReImaGin outperforms both text-only reasoning and specialist vision-tool baselines, achieving up to a 25% improvement.
GenDSR is a framework that transfers intermediate representations from video generation models (VGMs) into vision‑language models (VLMs) to improve dynamic spatial reasoning. It uses two complementary branches—a raw branch that preserves holistic generative context and a delta branch that emphasizes temporal variation—and fuses them with task‑conditioned token‑wise gates before injecting the result into the native visual representation. On the DSR‑Bench benchmark, GenDSR achieves a 67.2% score, surpassing previous methods and setting a new state‑of‑the‑art.
arXiv:2606. 16783v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) excel at visual reasoning but rely on text-based chain-of-thought (CoT), lacking interpretable visual intermediates.
arXiv:2608. 05242v1 Announce Type: new Abstract: In this work, we explore an alternative paradigm for spatial reasoning by explicitly disentangling 3D perception from reasoning, rather than jointly acquiring implicit 3D perception and reasoning through large-scale training.
arXiv:2609.24813v1 Announce Type: new Abstract: Vision-Language Models (VLMs) have demonstrated remarkable capabilities in multimodal tasks, yet they still exhibit poor ability in spatial reasoning....
arXiv:2511. 07403v2 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) have achieved remarkable progress in vision-language tasks, but continue to struggle with spatial reasoning.
In this work, we explore an alternative paradigm for spatial reasoning by explicitly disentangling 3D perception from reasoning, rather than jointly acquiring implicit 3D perception and reasoning through large-scale training. Our key observation is that modern perception models excel at estimating continuous 3D geometry, whereas large language models (LLMs) are particularly effective at compositional and symbolic reasoning.