pith. sign in

arxiv: 2606.17539 · v1 · pith:MKTPSXFCnew · submitted 2026-06-16 · 💻 cs.CV · cs.AI

Reinforcing Dual-Path Reasoning in Spatial Vision Language Models

Yatai Ji , An-Chieh Cheng , Yang Fu , Yukang Chen , Han Zhang , Zhaojing Yang , Wei Huang , Ka Chun Cheung
show 8 more authors
Song Han Vidya Nariyambut Murali Pavlo Molchanov Jan Kautz Simon See Hongxu Yin Ping Luo Sifei Liu
This is my paper

Pith reviewed 2026-06-27 01:39 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords spatial vision language modelsreinforcement learningdual-path reasoning3D localizationchain-of-thoughtspatial benchmarksmutual reinforcement
0
0 comments X

The pith

Reinforcement learning trains one spatial VLM to use both language deduction and 3D detection paths

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that spatial vision language models can be improved by giving them two different ways to reason about space. One way uses only language for step-by-step deduction while the other first detects 3D object positions before making calculations. A cold-start supervised stage creates training examples for both paths, then reinforcement learning optimizes the model using accuracy and detection rewards. The result is a single model that performs better than previous spatial VLMs on multiple benchmarks. The two paths reinforce each other and the model works across new datasets without task-specific changes.

Core claim

SR-REAL equips a spatial VLM with Language-Only Reasoning for pure linguistic deduction and Detect-Then-Reason for first locating 3D cues like centers or boxes then performing geometric inference. After blended cold-start supervision and RL with accuracy, format, and discrete center-based detection rewards, the model supports both paths in one policy. DTR excels at region-aware tasks through precise 3D localization while LOR strengthens general spatial reasoning, joint training produces mutual reinforcement and positive transfer, and the model generalizes across datasets and domains without per-task tuning.

What carries the argument

Dual-path framework with Language-Only Reasoning (LOR) for step-by-step linguistic deduction and Detect-Then-Reason (DTR) for 3D geometric cue detection via region tokens before explicit inference, optimized by RL after blended cold-start SFT

Load-bearing premise

High-quality blended cold-start data for LOR and DTR chain-of-thought supervision can be reliably constructed and the accuracy, format, and detection rewards will produce stable RL optimization without interference between paths or need for per-task tuning.

What would settle it

An experiment showing that a model trained with this dual-path RL procedure fails to outperform single-path spatial VLM baselines on held-out benchmarks or cannot stably support both LOR and DTR without interference would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.17539 by An-Chieh Cheng, Han Zhang, Hongxu Yin, Jan Kautz, Ka Chun Cheung, Pavlo Molchanov, Ping Luo, Sifei Liu, Simon See, Song Han, Vidya Nariyambut Murali, Wei Huang, Yang Fu, Yatai Ji, Yukang Chen, Zhaojing Yang.

Figure 1
Figure 1. Figure 1: A spatial imagination query where SR-ReaL resolves the task under both reasoning paths—Language-Only Reasoning (LOR) and Detect-Then-Reason (DTR) (right-most column). Prior art (left-most column) fail on such examples due to inaccurate or insufficient geometric deduction. 1. Introduction Large Vision-Language Models (VLMs) have rapidly advanced in interpreting and reasoning over visual content, driven by i… view at source ↗
Figure 2
Figure 2. Figure 2: Two-stage Training Pipeline: In Stage 1 (cold-start SFT), we fine-tune the spatial VLM with spatial CoT, 2D/3D grounding (global and region prompts), and region-prompted multimodal QA to initialize spatial reasoning ability. In Stage 2, we apply RL on multiple-choice and filling spatial QA, optimizing grouped rollouts of LOR/DTR trajectories with accuracy, format, and 3D-center detection rewards. Spatial R… view at source ↗
Figure 3
Figure 3. Figure 3: CoT Data Construction: We generate step-by-step CoT of two spatial reasoning paths: Language-Only Reasoning (top), Detect-Then-Reason with geometry-grounded deduction (middle). Complex Spatial tasks Construction: Using multimodal scene-graph datasets that provide both visual and geometric annotations, we prompt LVLM to generate higher-level reasoning task data (bottom). region prompt is marked as <mask> in… view at source ↗
Figure 4
Figure 4. Figure 4: Visualization examples of our model. On the fundamental spatial question (spatial relationship and [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
read the original abstract

Spatial VLMs have made substantial progress in geometric perception, yet complex spatial reasoning requiring multi-step inference over depth, distance, and scene relations remains challenging. Moreover, different spatial queries call for fundamentally different strategies: some are best addressed through purely linguistic, step-by-step deduction, while others require explicit 3D grounding before quantitative inference. We present Dual-Path Spatial Reasoning via Reinforcement Learning for Spatial VLMs (SR-REAL), a unified framework that equips a spatial VLM with two complementary reasoning paths: Language-Only Reasoning (LOR), which performs step-by-step linguistic deduction, and Detect-Then-Reason (DTR), which detects 3D geometric cues (e.g., centers or bounding boxes) via region tokens before explicit geometric inference. SR-REAL begins with a cold-start supervised fine-tuning stage that constructs LOR and DTR chain-of-thought supervision and exposes a region-to-3D interface, followed by RL that optimizes the policy model with accuracy and format rewards; for DTR, a discrete center-based detection reward further refines geometric alignment. Across diverse spatial benchmarks, SR-REAL significantly outperforms spatial VLM baselines: (i) a single RL-trained model supports both reasoning paths, with DTR excelling in region-aware tasks through precise 3D localization and LOR enhancing general spatial reasoning; (ii) jointly training both paths fosters mutual reinforcement; (iii) high-quality, blended cold-start data is crucial for stable RL optimization; and (iv) the model generalizes across datasets and domains without per-task tuning, demonstrating positive transfer between LOR and DTR.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces SR-REAL, a framework for spatial vision-language models that equips a single policy with two complementary reasoning paths—Language-Only Reasoning (LOR) via step-by-step linguistic deduction and Detect-Then-Reason (DTR) via region-token-based 3D localization—trained first with blended cold-start SFT supervision and then with RL using accuracy, format, and discrete center-based detection rewards. It claims that the resulting model outperforms spatial VLM baselines across benchmarks, that joint training produces mutual reinforcement between paths, that high-quality blended cold-start data is essential for stable optimization, and that the model generalizes across datasets without per-task tuning.

Significance. If the empirical results and ablations hold, the work would demonstrate a practical route to flexible spatial reasoning in VLMs by showing that a single RL-trained model can maintain both linguistic and grounded paths with positive transfer, addressing the limitation that different spatial queries require qualitatively different inference strategies.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'a single RL-trained model supports both reasoning paths' with 'mutual reinforcement' and 'no per-task tuning' rests on the unshown construction of blended LOR/DTR chain-of-thought traces and the choice of reward weights; without equations or pseudocode describing the mixing procedure, region-token exposure, or reward combination, it is impossible to evaluate whether path interference was avoided or whether the reported positive transfer is reproducible.
  2. [Abstract] Abstract: the statement that 'high-quality, blended cold-start data is crucial for stable RL optimization' is presented as a key finding, yet the manuscript provides no ablation on data quality, blending ratios, or the accuracy/format/detection reward formulation; this directly underpins claims (ii) and (iii) and must be substantiated with concrete construction details and sensitivity results.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for highlighting the need for greater methodological transparency. We agree that explicit details on data construction, mixing procedures, and reward formulations are required for reproducibility and will incorporate them in the revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'a single RL-trained model supports both reasoning paths' with 'mutual reinforcement' and 'no per-task tuning' rests on the unshown construction of blended LOR/DTR chain-of-thought traces and the choice of reward weights; without equations or pseudocode describing the mixing procedure, region-token exposure, or reward combination, it is impossible to evaluate whether path interference was avoided or whether the reported positive transfer is reproducible.

    Authors: We acknowledge the concern. The current manuscript describes the overall pipeline at a high level but does not provide the requested equations or pseudocode. In the revised version we will add a new subsection (and appendix) containing: (1) the exact procedure and ratio used to blend LOR and DTR CoT traces during cold-start SFT, (2) the region-token exposure mechanism and its integration with the vision encoder, and (3) the full reward function with explicit weights for accuracy, format, and discrete center-based detection terms. These additions will allow direct assessment of path interference and reproducibility of the reported transfer effects. revision: yes

  2. Referee: [Abstract] Abstract: the statement that 'high-quality, blended cold-start data is crucial for stable RL optimization' is presented as a key finding, yet the manuscript provides no ablation on data quality, blending ratios, or the accuracy/format/detection reward formulation; this directly underpins claims (ii) and (iii) and must be substantiated with concrete construction details and sensitivity results.

    Authors: We agree that the claim requires empirical support. The present version reports only the final performance after using the blended data and does not contain the requested ablations. We will add a new experimental subsection with: (a) sensitivity results across different blending ratios, (b) comparisons of high- versus lower-quality cold-start data, and (c) ablations on the individual reward components and their relative weights. These results will be used to substantiate claims (ii) and (iii) regarding mutual reinforcement and the necessity of high-quality blended supervision. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical RL outcomes with independent benchmark results

full rationale

The paper describes an RL training pipeline (cold-start SFT followed by accuracy/format/detection rewards) and reports empirical gains on spatial benchmarks. No equations, uniqueness theorems, or derivations are invoked that reduce any claimed result to a fitted parameter or self-citation by construction. The construction of blended LOR/DTR supervision and choice of rewards are presented as methodological inputs whose effectiveness is measured externally on held-out benchmarks; they do not constitute a self-definitional loop or renamed known result. The work is therefore self-contained against external evaluation and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no explicit free parameters, axioms, or invented entities beyond standard components of VLM fine-tuning and RL; all elements appear drawn from existing techniques.

pith-pipeline@v0.9.1-grok · 5878 in / 1194 out tokens · 52707 ms · 2026-06-27T01:39:45.694981+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

101 extracted references · 1 canonical work pages

  1. [2]

    Qwen2.5-vl technical report

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Ming - Hsuan Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, and Junyang Lin. Qwen2.5-vl technical repo...

    Pith/arXiv arXiv 2025

  2. [3]

    Omni3d: A large benchmark and model for 3d object detection in the wild

    Garrick Brazil, Abhinav Kumar, Julian Straub, Nikhila Ravi, Justin Johnson, and Georgia Gkioxari. Omni3d: A large benchmark and model for 3d object detection in the wild. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. URL https://github.com/facebookresearch/omni3d. Code and dataset release: facebookresea...

    arXiv 2023

  3. [9]

    Rynnbrain: Open embodied foundation models

    Ronghao Dang, Jiayan Guo, Bohan Hou, Sicong Leng, Kehan Li, Xin Li, Jiangpin Liu, Yunxuan Mao, Zhikai Wang, Yuqian Yuan, Minghao Zhu, Xiao Lin, Yang Bai, Qian Jiang, Yaxi Zhao, Minghua Zeng, Junlong Gao, Yuming Jiang, Jun Cen, Siteng Huang, Liuyi Wang, Wenqiao Zhang, Chengju Liu, Jianfei Yang, Shijian Lu, and Deli Zhao. Rynnbrain: Open embodied foundation...

    arXiv 2026

  4. [11]

    Embspatial-bench: Benchmarking spatial understanding for embodied tasks with large vision-language models

    Mengfei Du, Binhao Wu, Zejun Li, Xuanjing Huang, and Zhongyu Wei. Embspatial-bench: Benchmarking spatial understanding for embodied tasks with large vision-language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 2024

    2024

  5. [13]

    Gemini: A family of highly capable multimodal models, 2023

    Google DeepMind . Gemini: A family of highly capable multimodal models, 2023. URL https://deepmind.google/gemini/

    2023

  6. [16]

    Seeing what you miss: Vision-language pre-training with semantic completion learning

    Yatai Ji, Rongcheng Tu, Jie Jiang, Weijie Kong, Chengfei Cai, Wenzhe Zhao, Hongfa Wang, Yujiu Yang, and Wei Liu. Seeing what you miss: Vision-language pre-training with semantic completion learning. In CVPR , pages 6789--6798. IEEE , 2023

    2023

  7. [17]

    IDA-VLM: towards movie understanding via id-aware large vision-language model

    Yatai Ji, Shilong Zhang, Jie Wu, Peize Sun, Weifeng Chen, Xuefeng Xiao, Sidi Yang, Yujiu Yang, and Ping Luo. IDA-VLM: towards movie understanding via id-aware large vision-language model. In ICLR . OpenReview.net, 2025

    2025

  8. [18]

    What’s "up" with vision-language models? investigating their struggle with spatial reasoning

    Amita Kamath, Jack Hessel, and Kai-Wei Chang. What’s "up" with vision-language models? investigating their struggle with spatial reasoning. In EMNLP, 2023

    2023

  9. [19]

    Referitgame: Referring to objects in photographs of natural scenes

    Sahar Kazemzadeh, Vicente Ordonez, Mark Matten, and Tamara Berg. Referitgame: Referring to objects in photographs of natural scenes. GitHub repository: https://github.com/lichengunc/refer, 2014

    2014

  10. [20]

    Omninocs: A unified NOCS dataset and model for 3d lifting of 2d objects

    Akshay Krishnan, Abhijit Kundu, Kevis - Kokitsi Maninis, James Hays, and Matthew Brown. Omninocs: A unified NOCS dataset and model for 3d lifting of 2d objects. In ECCV (75) , volume 15133 of Lecture Notes in Computer Science, pages 127--145. Springer, 2024 a

    2024

  11. [21]

    Omninocs: A unified nocs dataset and model for 3d lifting of 2d objects

    Akshay Krishnan, Abhijit Kundu, Kevis-Kokitsi Maninis, James Hays, and Matthew Brown. Omninocs: A unified nocs dataset and model for 3d lifting of 2d objects. In European Conference on Computer Vision, pages 127--145. Springer, 2024 b

    2024

  12. [26]

    Vila: On pre-training for visual language models

    Ji Lin, Hongxu Yin, Wei Ping, Pavlo Molchanov, Mohammad Shoeybi, and Song Han. Vila: On pre-training for visual language models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 26689--26699, 2024

    2024

  13. [28]

    Visual spatial reasoning

    Fangyu Liu, Guy Emerson, and Nigel Collier. Visual spatial reasoning. Transactions of the Association for Computational Linguistics, 11: 0 635--651, 2023

    2023

  14. [31]

    Spatialcot: Advancing spatial reasoning through coordinate alignment and chain-of-thought for embodied task planning, 2025 b

    Yuecheng Liu, Dafeng Chi, Shiguang Wu, Zhanguang Zhang, Yaochen Hu, Lingfeng Zhang, Yingxue Zhang, Shuang Wu, Tongtong Cao, Guowei Huang, Helong Huang, Guangjian Tian, Weichao Qiu, Xingyue Quan, Jianye Hao, and Yuzheng Zhuang. Spatialcot: Advancing spatial reasoning through coordinate alignment and chain-of-thought for embodied task planning, 2025 b . URL...

    arXiv 2025

  15. [32]

    Nvila: Efficient frontier visual language models

    Zhijian Liu, Ligeng Zhu, Baifeng Shi, Zhuoyang Zhang, Yuming Lou, Shang Yang, Haocheng Xi, Shiyi Cao, Yuxian Gu, Dacheng Li, et al. Nvila: Efficient frontier visual language models. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 4122--4134, 2025 c

    2025

  16. [34]

    Spatialllm: A compound 3d-informed design towards spatially-intelligent large multimodal models

    Wufei Ma, Luoxin Ye, Celso de Melo, Alan L Yuille, and Jieneng Chen. Spatialllm: A compound 3d-informed design towards spatially-intelligent large multimodal models. In CVPR, 2025 b

    2025

  17. [37]

    Gpt-4o, 2024

    OpenAI . Gpt-4o, 2024. URL https://openai.com

    2024

  18. [40]

    Learning to localize objects improves spatial reasoning in visual-llms

    Kanchana Ranasinghe, Satya Narayan Shukla, Omid Poursaeed, Michael S Ryoo, and Tsung-Yu Lin. Learning to localize objects improves spatial reasoning in visual-llms. In CVPR, pages 12977--12987, 2024

    2024

  19. [41]

    Plummer, Ranjay Krishna, Kuo-Hao Zeng, and Kate Saenko

    Arijit Ray, Jiafei Duan, Ellis Brown, Reuben Tan, Dina Bashkirova, Rose Hendrix, Kiana Ehsani, Aniruddha Kembhavi, Bryan A. Plummer, Ranjay Krishna, Kuo-Hao Zeng, and Kate Saenko. Sat: Spatial aptitude training for multimodal language models, 2025

    2025

  20. [43]

    An empirical analysis on spatial reasoning capabilities of large multimodal models

    Fatemeh Shiri, Xiao-Yu Guo, Mona Far, Xin Yu, Reza Haf, and Yuan-Fang Li. An empirical analysis on spatial reasoning capabilities of large multimodal models. In EMNLP, 2024

    2024

  21. [44]

    Sparkle: Mastering basic spatial capabilities in vision language models elicits generalization to spatial reasoning, 2025

    Yihong Tang, Ao Qu, Zhaokai Wang, Dingyi Zhuang, Zhaofeng Wu, Wei Ma, Shenhao Wang, Yunhan Zheng, Zhan Zhao, and Jinhua Zhao. Sparkle: Mastering basic spatial capabilities in vision language models elicits generalization to spatial reasoning, 2025. URL https://arxiv.org/abs/2410.16162

    arXiv 2025

  22. [45]

    Qwen3-vl technical report

    Qwen Team. Qwen3-vl technical report. CoRR, abs/2511.21631, 2025

    Pith/arXiv arXiv 2025

  23. [47]

    Cambrian-1: A fully open, vision-centric exploration of multimodal llms

    Peter Tong, Ellis Brown, Penghao Wu, Sanghyun Woo, Adithya Jairam Vedagiri IYER, Sai Charitha Akula, Shusheng Yang, Jihan Yang, Manoj Middepogu, Ziteng Wang, et al. Cambrian-1: A fully open, vision-centric exploration of multimodal llms. Advances in Neural Information Processing Systems, 37: 0 87310--87356, 2024

    2024

  24. [50]

    Is a picture worth a thousand words? delving into spatial reasoning for vision language models

    Jiayu Wang, Yifei Ming, Zhenmei Shi, Vibhav Vineet, Xin Wang, Sharon Li, and Neel Joshi. Is a picture worth a thousand words? delving into spatial reasoning for vision language models. Advances in Neural Information Processing Systems, 37: 0 75392--75421, 2024 a

    2024

  25. [51]

    Embodiedscan: A holistic multi-modal 3d perception suite towards embodied AI

    Tai Wang, Xiaohan Mao, Chenming Zhu, Runsen Xu, Ruiyuan Lyu, Peisen Li, Xiao Chen, Wenwei Zhang, Kai Chen, Tianfan Xue, Xihui Liu, Cewu Lu, Dahua Lin, and Jiangmiao Pang. Embodiedscan: A holistic multi-modal 3d perception suite towards embodied AI . In CVPR , pages 19757--19767. IEEE , 2024 b

    2024

  26. [52]

    Wu et al

    Q. Wu et al. Visual spatial tuning: Bootstrapping spatial intelligence in vlms via 3d-aware perception and reinforcement learning. arXiv preprint arXiv:25xx.xxxxx, 2025

    2025

  27. [53]

    Realworldqa: A benchmark for real-world spatial understanding

    xAI . Realworldqa: A benchmark for real-world spatial understanding. https://huggingface.co/datasets/xai-org/RealworldQA, 2024. Accessed: 2025-11-13

    2024

  28. [54]

    Visual spatial tuning

    Rui Yang, Ziyu Zhu, Yanwei Li, Jingjia Huang, Shen Yan, Siyuan Zhou, Zhe Liu, Xiangtai Li, Shuangye Li, Wenqian Wang, Yi Lin, and Hengshuang Zhao. Visual spatial tuning. CoRR, abs/2511.05491, 2025 a

    arXiv 2025

  29. [59]

    Video-3d llm: Learning position-aware video representation for 3d scene understanding

    Duo Zheng, Shijia Huang, and Liwei Wang. Video-3d llm: Learning position-aware video representation for 3d scene understanding. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 8995--9006, 2025

    2025

  30. [61]

    FirstName LastName , title =

  31. [62]

    FirstName Alpher , title =

  32. [63]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  33. [64]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  34. [65]

    FirstName Alpher and FirstName Gamow , title =

  35. [66]

    EMNLP , year=

    What’s "up" with vision-language models? Investigating their struggle with spatial reasoning , author=. EMNLP , year=

  36. [67]

    CoRR , volume =

    Rui Yang and Ziyu Zhu and Yanwei Li and Jingjia Huang and Shen Yan and Siyuan Zhou and Zhe Liu and Xiangtai Li and Shuangye Li and Wenqian Wang and Yi Lin and Hengshuang Zhao , title =. CoRR , volume =

  37. [68]

    CoRR , volume =

    Gemini Team , title =. CoRR , volume =

  38. [69]

    CoRR , volume =

    Ronghao Dang and Jiayan Guo and Bohan Hou and Sicong Leng and Kehan Li and Xin Li and Jiangpin Liu and Yunxuan Mao and Zhikai Wang and Yuqian Yuan and Minghao Zhu and Xiao Lin and Yang Bai and Qian Jiang and Yaxi Zhao and Minghua Zeng and Junlong Gao and Yuming Jiang and Jun Cen and Siteng Huang and Liuyi Wang and Wenqiao Zhang and Chengju Liu and Jianfei...

  39. [70]

    Qwen2.5-VL Technical Report , journal =

    Shuai Bai and Keqin Chen and Xuejing Liu and Jialin Wang and Wenbin Ge and Sibo Song and Kai Dang and Peng Wang and Shijie Wang and Jun Tang and Humen Zhong and Yuanzhi Zhu and Ming. Qwen2.5-VL Technical Report , journal =

  40. [71]

    CoRR , volume =

    Qwen Team , title =. CoRR , volume =

  41. [72]

    Transactions of the Association for Computational Linguistics , volume=

    Visual spatial reasoning , author=. Transactions of the Association for Computational Linguistics , volume=. 2023 , publisher=

    2023

  42. [73]

    arXiv preprint arXiv:2308.09778 , year=

    Towards grounded visual spatial reasoning in multi-modal vision language models , author=. arXiv preprint arXiv:2308.09778 , year=

    arXiv

  43. [74]

    CVPR , pages=

    Learning to localize objects improves spatial reasoning in visual-LLMs , author=. CVPR , pages=

  44. [75]

    EMNLP , year=

    An empirical analysis on spatial reasoning capabilities of large multimodal models , author=. EMNLP , year=

  45. [76]

    arXiv preprint arXiv:2504.17207 , year=

    Perspective-aware reasoning in vision-language models via mental imagery simulation , author=. arXiv preprint arXiv:2504.17207 , year=

    arXiv

  46. [77]

    Advances in Neural Information Processing Systems , volume=

    Is a picture worth a thousand words? delving into spatial reasoning for vision language models , author=. Advances in Neural Information Processing Systems , volume=

  47. [78]

    2025 , eprint=

    Sparkle: Mastering Basic Spatial Capabilities in Vision Language Models Elicits Generalization to Spatial Reasoning , author=. 2025 , eprint=

    2025

  48. [79]

    2025 , eprint=

    SpatialCoT: Advancing Spatial Reasoning through Coordinate Alignment and Chain-of-Thought for Embodied Task Planning , author=. 2025 , eprint=

    2025

  49. [80]

    arXiv preprint arXiv:2505.12448 , year=

    SSR: Enhancing depth perception in vision-language models via rationale-guided spatial reasoning , author=. arXiv preprint arXiv:2505.12448 , year=

    arXiv

  50. [81]

    CVPR , year=

    SpatialLLM: A compound 3D-informed design towards spatially-intelligent large multimodal models , author=. CVPR , year=

  51. [82]

    arXiv preprint arXiv:2504.20024 , year=

    SpatialReasoner: Towards explicit and generalizable 3D spatial reasoning , author=. arXiv preprint arXiv:2504.20024 , year=

    arXiv

  52. [83]

    arXiv preprint arXiv:2401.12168 , year =

    SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities , author =. arXiv preprint arXiv:2401.12168 , year =

    arXiv

  53. [84]

    arXiv preprint arXiv:2403.02330 , year =

    RegionGPT: Towards Region Understanding Vision Language Model , author =. arXiv preprint arXiv:2403.02330 , year =

    arXiv

  54. [85]

    arXiv preprint arXiv:2406.01584 , year =

    SpatialRGPT: Grounded Spatial Reasoning in Vision Language Models , author =. arXiv preprint arXiv:2406.01584 , year =

    arXiv

  55. [86]

    arXiv preprint arXiv:2409.18125 , year =

    LLaVA-3D: A Simple yet Effective Pathway to Empowering LMMs with 3D-awareness , author =. arXiv preprint arXiv:2409.18125 , year =

    arXiv

  56. [87]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Video-3d llm: Learning position-aware video representation for 3d scene understanding , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  57. [88]

    arXiv preprint arXiv:2509.13317 , year =

    3D Aware Region Prompted Vision Language Model , author =. arXiv preprint arXiv:2509.13317 , year =

    arXiv

  58. [89]

    arXiv preprint arXiv:2501.19393 , year =

    s1: Simple Test-Time Scaling , author =. arXiv preprint arXiv:2501.19393 , year =

    Pith/arXiv arXiv

  59. [90]

    arXiv preprint arXiv:2501.06186 , year =

    LlamaV-o1: Rethinking Step-by-step Visual Reasoning in LLMs , author =. arXiv preprint arXiv:2501.06186 , year =

    arXiv

  60. [91]

    arXiv preprint arXiv:2501.12948 , year =

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning , author =. arXiv preprint arXiv:2501.12948 , year =

    Pith/arXiv arXiv

  61. [92]

    arXiv preprint arXiv:2503.06749 , year =

    Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models , author =. arXiv preprint arXiv:2503.06749 , year =

    Pith/arXiv arXiv

  62. [93]

    arXiv preprint arXiv:2503.07365 , year =

    MM-Eureka: Exploring Visual Aha Moment with Rule-based Large-scale Reinforcement Learning , author =. arXiv preprint arXiv:2503.07365 , year =

    Pith/arXiv arXiv

  63. [94]

    arXiv preprint arXiv:2503.10615 , year =

    R1-OneVision: Advancing Generalized Multimodal Reasoning through Cross-Modal Formalization , author =. arXiv preprint arXiv:2503.10615 , year =

    Pith/arXiv arXiv

  64. [95]

    arXiv preprint arXiv:2503.12937 , year =

    R1-VL: Learning to Reason with Multimodal Large Language Models via Step-wise Group Relative Policy Optimization , author =. arXiv preprint arXiv:2503.12937 , year =

    Pith/arXiv arXiv

  65. [96]

    arXiv preprint arXiv:2504.08837 , year =

    VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning , author =. arXiv preprint arXiv:2504.08837 , year =

    Pith/arXiv arXiv

  66. [97]

    arXiv preprint arXiv:2506.12322 , year =

    3D-R1: Reinforcing 3D Spatial Reasoning and Understanding in Large Multimodal Models , author =. arXiv preprint arXiv:2506.12322 , year =

    arXiv

  67. [98]

    arXiv preprint arXiv:25xx.xxxxx , year =

    Visual Spatial Tuning: Bootstrapping Spatial Intelligence in VLMs via 3D-Aware Perception and Reinforcement Learning , author =. arXiv preprint arXiv:25xx.xxxxx , year =

  68. [99]

    arXiv preprint arXiv:2510.08531 , year =

    SpatialLadder: Progressive Training for Spatial Reasoning in Vision--Language Models , author =. arXiv preprint arXiv:2510.08531 , year =

    arXiv

  69. [100]

    arXiv preprint arXiv:2505.23678 , year =

    Grounded Reinforcement Learning for Visual Reasoning , author =. arXiv preprint arXiv:2505.23678 , year =

    Pith/arXiv arXiv

  70. [101]

    arXiv preprint arXiv:2308.12966 , year =

    Qwen-VL: A Frontier Large Vision-Language Model with Versatile Abilities , author =. arXiv preprint arXiv:2308.12966 , year =

    Pith/arXiv arXiv

  71. [102]

    arXiv preprint arXiv:2305.05662 , year =

    InternVL: Scaling up Vision Foundation Models and Aligning for Generic Vision-Language Understanding , author =. arXiv preprint arXiv:2305.05662 , year =

    arXiv

  72. [103]

    arXiv preprint arXiv:2402.16835 , year =

    LLaVA-1.5: Improved Reasoning, Grounding, and Chart Understanding , author =. arXiv preprint arXiv:2402.16835 , year =

    arXiv

  73. [104]

    arXiv preprint arXiv:2312.07533 , year =

    VILA: Large-Scale Vision-Language Pre-Training with Region-Based Supervision , author =. arXiv preprint arXiv:2312.07533 , year =

    arXiv

  74. [105]

    2024 , url =

    GPT-4o , author =. 2024 , url =

    2024

  75. [106]

    2023 , url =

    Gemini: A Family of Highly Capable Multimodal Models , author =. 2023 , url =

    2023

  76. [107]

    arXiv preprint arXiv:2510.15870 , year=

    OmniVinci: Enhancing Architecture and Data for Omni-Modal Understanding LLM , author=. arXiv preprint arXiv:2510.15870 , year=

    arXiv

  77. [108]

    OmniNOCS:

    Akshay Krishnan and Abhijit Kundu and Kevis. OmniNOCS:

  78. [109]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Vila: On pre-training for visual language models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  79. [110]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Nvila: Efficient frontier visual language models , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  80. [111]

    arXiv preprint arXiv:2503.15558 , year =

    Cosmos-Reason1: From Physical Common Sense To Embodied Reasoning , author =. arXiv preprint arXiv:2503.15558 , year =

    Pith/arXiv arXiv

Showing first 80 references.