Think-Capture is a query‑conditioned framework that selectively acquires 3D geometry for task‑critical objects to improve spatial reasoning in Vision‑Language Models. It uses a 3D toolkit to lift zero‑shot masks into a common coordinate frame, a Symbolic 3D State Abstraction to encode object and camera states as structured text, and Hierarchical Group Relative Policy Optimization to train the VLM on when geometry helps or misleads. The resulting Think‑Capture‑4B model achieves significant gains on MindCube‑Tiny, MMSI‑Bench, and ViewSpatial benchmarks while remaining robust to perturbations and fitting on a single 24 GB GPU.
By Junjie Wang, Zhihao Yuan, Yunhan Tang, Sheng Xu, Jiaqi Mao, Xiaoyu Ye, Yueru Luo, Xu Yan, Chun-Mei Feng, Zhen Li
Vision-language models commonly project all tokens produced by a pretrained vision encoder into a large language model. However, final-layer features can discard text, local attributes, and spatial relationships, while high-resolution inputs substantially increase context length and inference latency.
arXiv:2609.08345v1 Announce Type: cross
Abstract: Representing a 3D scene as multi-view images allows 2D VLMs to reason in 3D by reusing priors from pre-training, sidestepping the scarcity of annotat...
By Nhat-Tan Bui, Varshini Elangovan, Arun Reddy Anugu, Sreyas Mohan, Wei Ye, Dilin Wang, JQ Huang, Rakesh Ranjan, Aviral Chharia, Fernando De la Torre
arXiv:2608. 01899v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) perform well on commonsense reasoning tasks but struggle with visual spatial reasoning.
By Jing Wu, Jianhua Wu, Jiayi Guan, Jiahong Chen, Jinghui Lu, Hangjun Ye, Bingzhao Gao, Long Chen
VETO (Video Efficient Token Optimization for Vision Language Models) is a plug‑in that reduces the quadratic cost of visual tokens in long‑video inference by applying dual‑axis compression: an intra‑frame compressor merges semantically similar tokens within each frame, and an inter‑frame compressor merges temporally redundant frames. By first compressing spatial dimensions, VETO lowers the cost of subsequent global temporal matching, surpassing single‑axis methods and achieving up to 45% faster inference on models such as LLaVA‑OneVision‑7B while maintaining or improving accuracy. The approach is universally applicable across LLaVA‑OneVision, InternVL‑2.5, and LongVA, preserving or enhancing zero‑shot accuracy even under extreme token budgets.
By Gueter Josmy Faure, Hao Ping Wang, Min-Hung Chen, Winston H. Hsu
arXiv:2609.06880v1 Announce Type: cross
Abstract: Reasoning over language instructions in embodied tasks such as robotics often requires understanding spatial relations from a speaker's situated pers...
By Mimo Shirasaka, Haochen Zhang, Yonatan Bisk
DK‑VLM introduces Dual‑Kernel Attention, where a visual token’s attention is computed from both its input position and its geometric position derived from 3D points and camera poses. A semantic kernel preserves pretrained attention while a geometric kernel modulates attention via point displacements and relative view transformations, with a learned gate fusing the two. The model also employs Reference‑Frame Composition supervision to update its reference frame across successive judgments, achieving top scores on six spatial benchmarks and strong performance on VSI‑Bench and ReVSI‑32.
By Han Su, Meng Tian, Tianyu Huang, Mingyu Gao, Lennox Wei, Dongxu Li, Jianhua Han, Hang Xu, Xiaohe Wu, Wangmeng Zuo
GeoGR introduces a two‑stage framework for efficient 3D vision‑language models. The first stage, GeoSemZip, compresses multi‑view visual tokens into compact 3D voxel representations while preserving important scene information through contextual anchors. The second stage, GroupRoute, dynamically routes tokens through the language model, allowing low‑relevance tokens to skip layers and rejoin later, thereby reducing computation without significant loss of performance.
By Yongxin Zheng, Hao Wu, Fang Li, Hu Fu, Peiran Yin, Xudong Wang, Haozhe Hu, Xinghao Chen, Yunpu Ma, Wei Zhang, Xiaoyu Shen
arXiv:2603. 06576v2 Announce Type: replace-cross Abstract: The integration of Large Language Models (LLMs) into autonomous driving has attracted growing interest for their strong reasoning and semantic understanding abilities, which are essential for handling complex decision-making and long-tail scenarios.
By Thomas Monninger, Shaoyuan Xie, Qi Alfred Chen, Sihao Ding
arXiv:2606. 19253v1 Announce Type: cross Abstract: Existing approaches to 3D scene understanding in Vision-Language Models (VLMs) either rely on complex, model-specific geometry encoders or large training budgets in pursuit of spatial reasoning.
By Bart{\l}omiej Baranowski, Dave Zhenyu Chen, Matthias Nie{\ss}ner
3D vision-language models (3D VLMs) enable spatial reasoning over multi-view scenes but suffer from substantial token redundancy due to duplicated observations and large uninformative regions, leading to high computational cost. Although visual token compression has shown promise in accelerating 2D VLMs, it fails to capture the structured nature of 3D scenes and leads to incomplete spatial coverage and loss of fine-grained details.
DART is a fully differentiable Dynamic Adaptive Region Tokenizer that replaces fixed‑grid tokenizers in vision models. It uses learnable region scores and quantile‑based partitioning to create content‑aware patches of varying sizes, allocating more tokens to information‑rich areas. In experiments, a DART‑equipped DeiT‑Small matches the performance of a larger DeiT‑Base while nearly doubling inference speed, and the approach also benefits dense prediction and video tasks.
By Shicheng Yin, Kaixuan Yin, Yang Liu, Weixing Chen, Liang Lin