SpatialReasoner-R1Fine-Grained Preference Optimization for Spatial Reasoning in VLMs

Yifan Shen^♦, Yuanzhe Liu^♣, Jingyuan Zhu^♣, Xu Cao^♦, Xiaofeng Zhang^♥,

Yixiao He^♦, Wenming Ye^♠, James Matthew Rehg^♦, Ismini Lourentzou^♦

^♦University of Illinois Urbana-Champaign^♣University of Pennsylvania^♥Shanghai Jiao Tong University^♠Google

NeurIPS 2025

TL;DR

We introduce SpatialReasoner-R1, a model for answering fine-grained spatial questions with interpretable, step-by-step reasoning. SpatialReasoner-R1 learns to connect descriptions and logical deductions to visible spatial evidence, helping models reason about object locations, relationships, and layouts.

SpatialReasoner-R1 example with visually grounded multi-step spatial reasoning — Fine-grained preference optimization for spatial reasoning

SpatialReasoner-R1 connects descriptions and reasoning to the objects and relationships visible in the scene.

Abstract

Current Vision-Language Models (VLMs) struggle with fine-grained spatial reasoning, particularly when multi-step logic and precise spatial alignment are required. In this work, we introduce SpatialReasoner-R1 a vision-language reasoning model designed to address these limitations. To construct high-quality supervision for spatial reasoning, we design a Multi-Model Monte Carlo Tree Search (M3CTS) method that generates diverse, logically consistent Long Chain-of-Thought (LongCoT) reasoning trajectories. In addition, we propose fine-grained Direct Preference Optimization (fDPO), which introduces segment-specific preference granularity for descriptive grounding and logical reasoning, guided by a spatial reward mechanism that evaluates candidate responses based on visual consistency, spatial grounding, and logical coherence. Experimental results demonstrate that fDPO achieves an average improvement of 4.1% over standard DPO across spatial quality tasks, and a 9.0% gain in spatial quantity tasks. SpatialReasoner-R1, trained with fDPO, sets a new SoTA on SpatialRGPT-Bench, outperforming the strongest baseline by 9.4% in average accuracy, while maintaining competitive performance on general vision-language tasks.

Contributions

SpatialReasoner-R1. We introduce SpatialReasoner-R1, a VLM designed for fine-grained LongCoT spatial reasoning that effectively generates interpretable, step-by-step explanations directly from 2D images. SpatialReasoner-R1 establishes a new SoTA in spatial understanding tasks, while maintaining robust performance on general vision-language benchmarks.
Fine-grained Direct Preference Optimization. To enhance training stability and precision, we propose a new fine-grained Direct Preference Optimization fDPO method that employs segment-specific learning updates tailored explicitly for descriptive grounding and logical reasoning.
Multi-Model Monte Carlo Tree Search. To address the scarcity of high-quality spatial reasoning data, we introduce a data generation pipeline that combines Multi-Model Monte Carlo Tree Search (M3CTS) with fine-grained spatial rewards, enabling the creation of diverse, logically consistent LongCoT trajectories for fine-grained preference training.

SpatialReasoner-R1 Method

SpatialReasoner-R1 takes an image, a question, and visual region prompts to generate a grounded description and multi-step spatial reasoning. Its training pipeline first constructs reasoning trajectories, then selects preference pairs, and finally optimizes the descriptive and reasoning segments separately.

SpatialReasoner-R1 architecture and three-stage training pipeline with M3CTS, preference-pair construction, and fine-grained DPO — M3CTS produces reasoning paths. Spatial rewards select positive and negative responses, and fDPO applies distinct preference updates to descriptive grounding and logical reasoning.

Generate reasoning paths with M3CTS

Multi-Model Monte Carlo Tree Search explores diverse long reasoning trajectories grounded in the scene.
Construct fine-grained preference pairs

Candidate responses receive separate descriptive, spatial, and reasoning scores. Reward-based selection produces preferences for the relevant response segments.
Optimize descriptions and reasoning with fDPO

Segment-specific preference updates improve visual grounding and logical coherence instead of treating the entire response as one undifferentiated preference.

Quantitative Results

SpatialRGPT-Bench evaluates qualitative relations and quantitative measurements. The two tables report each question type separately.

Qualitative spatial relations

Accuracy on relative-position and relative-size questions in SpatialRGPT-Bench.

Qualitative spatial relations
Model	Below/Above	Left/Right	Big/Small	Tall/Short	Wide/Thin	Behind/Front	Qual. accuracy
Gemini 2.0 Flash	58.33	68.57	16.98	50.00	15.38	53.63	44.29
Llama 4 Maverick	54.17	61.90	33.02	50.89	25.96	55.45	47.18
Gemini 1.5 Pro	85.83	56.19	58.49	71.42	55.76	60.00	65.14
ChatGPT-4o	87.50	80.00	53.77	63.39	51.92	60.90	66.67
SpatialBot-3B	52.50	62.86	57.54	49.11	49.04	62.73	55.56
SpaceThinker Qwen2.5VL-3B	89.16	63.81	76.41	56.25	56.73	70.91	69.25
InternVL2.5-78B	94.16	94.28	64.15	65.17	55.76	58.18	72.29
Sa2va 4B	22.50	25.71	25.47	16.07	27.88	30.91	24.65
Sa2va 8B	50.00	39.04	45.28	26.78	45.19	53.63	43.37
SpatialRGPT-8B	99.17	100.00	84.90	89.28	91.34	90.90	92.69
SpatialReasoner-R1 SFT 4B	79.16	78.09	55.66	66.96	59.61	75.45	69.41
SpatialReasoner-R1 SFT 8B	81.66	81.90	75.47	75.89	79.80	83.63	79.75
SpatialReasoner-R1 DPO 4B	91.66	91.42	69.81	65.17	71.15	85.45	79.29
SpatialReasoner-R1 DPO 8B	94.16	93.33	89.62	90.18	88.64	92.27	91.48
SpatialReasoner-R1 fDPO 4B	95.83	93.33	83.96	74.10	87.50	89.09	87.37
SpatialReasoner-R1 fDPO 8B	98.33	98.10	95.28	96.43	91.34	93.64	95.59

Quantitative spatial measurements

Accuracy on distance, width, height, and direction questions in SpatialRGPT-Bench.

Quantitative spatial measurements
Model	Direct distance	Horizontal distance	Vertical distance	Width	Height	Direction	Quant. accuracy
Gemini 2.0 Flash	9.45	10.65	26.41	10.52	30.82	54.20	22.43
Llama 4 Maverick	24.48	28.68	34.28	35.71	44.61	58.09	36.72
Gemini 1.5 Pro	14.18	17.21	14.15	19.54	36.09	30.84	21.90
ChatGPT-4o	—	—	—	—	—	60.75	—
SpatialBot-3B	6.00	15.51	8.00	10.52	18.75	39.00	15.62
SpaceThinker Qwen2.5VL-3B	24.32	17.21	59.43	23.27	23.62	32.35	28.97
InternVL2.5-78B	27.70	22.13	41.50	29.32	34.58	62.61	35.25
Sa2va 4B	13.51	15.57	19.81	13.53	12.03	10.28	14.02
Sa2va 8B	14.18	14.75	9.43	14.28	19.54	14.18	14.55
SpatialRGPT-8B	45.90	68.00	56.60	48.90	61.70	95.30	61.42
SpatialReasoner-R1 SFT 4B	22.29	27.86	31.13	25.56	33.80	47.66	30.71
SpatialReasoner-R1 SFT 8B	28.43	20.49	44.05	33.59	51.63	46.72	37.12
SpatialReasoner-R1 DPO 4B	47.97	46.72	60.37	45.11	55.63	91.58	56.61
SpatialReasoner-R1 DPO 8B	62.83	56.55	60.37	70.45	68.42	93.45	68.22
SpatialReasoner-R1 fDPO 4B	60.13	59.01	71.70	65.41	57.89	92.52	66.76
SpatialReasoner-R1 fDPO 8B	70.95	72.13	74.52	80.45	74.43	94.39	77.30

Qualitative Results

BibTeX

@inproceedings{shen2025finegrainedpreferenceoptimizationimproves,
  title={Fine-Grained Preference Optimization Improves Spatial Reasoning in VLMs},
  author={Shen, Yifan and Liu, Yuanzhe and Zhu, Jingyuan and Cao, Xu and Zhang, Xiaofeng and He, Yixiao and Ye, Wenming and Rehg, James Matthew and Lourentzou, Ismini},
  booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
  year={2025}
}

Template acknowledgements

This site is built upon the work of Nerfies, made available under the Creative Commons Attribution-ShareAlike 4.0 International License.