SpatialBlock: Enhancing Spatial Intelligence in LVLMs via Synthetic Block-Stacking Problem

Hugging Face Daily Papers Papers
spatial-intelligence lvlm synthetic-data 3d-reasoning block-stacking ai-research

Summary

This paper introduces SpatialBlock-15k, a synthetic dataset for block-stacking problems, to enhance 3D spatial reasoning in large vision-language models, demonstrating improved performance and generalization to real-world tasks.

Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this gap by using real-scene spatial question answering datasets that require dense geometric annotations. However, constructing such labels is costly, time-consuming, and often noisy due to reliance on external perception modules. In this work, we propose a novel paradigm inspired by human cognitive development: learning foundational spatial skills through structured block-manipulation tasks. We introduce SpatialBlock-15k, a synthetic dataset of 15,000 block-stacking problems covering 3D-to-2D projection, viewpoint transformation, and structural combination. The dataset further incorporates controlled color modulation as visual cues to encourage anchor-based reasoning in visually complex conditions. Experiments demonstrate that LVLMs trained on our dataset through either direct answering or reasoning-based prediction significantly outperform baselines and generalize to real-world spatial tasks, despite the dataset's synthetic and compact nature. Code and data are available at https://github.com/rsoohyun/SpatialBlock.
Original Article
View Cached Full Text

Cached at: 09/11/26, 06:16 AM

Paper page - SpatialBlock: Enhancing Spatial Intelligence in LVLMs via Synthetic Block-Stacking Problem

Source: https://huggingface.co/papers/2609.07064

Abstract

Large vision-language models trained on synthetic block-manipulation tasks improve 3D spatial reasoning and generalize to real-world visual tasks.

Large Vision-Language Models(LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to asspatial intelligence-- remains limited. Existing approaches attempt to address this gap by using real-scene spatial question answering datasets that require dense geometric annotations. However, constructing such labels is costly, time-consuming, and often noisy due to reliance on external perception modules. In this work, we propose a novel paradigm inspired by human cognitive development: learning foundational spatial skills through structured block-manipulation tasks. We introduce SpatialBlock-15k, a synthetic dataset of 15,000 block-stacking problems covering3D-to-2D projection,viewpoint transformation, andstructural combination. The dataset further incorporates controlled color modulation as visual cues to encourageanchor-based reasoningin visually complex conditions. Experiments demonstrate that LVLMs trained on our dataset through either direct answering orreasoning-based predictionsignificantly outperform baselines and generalize to real-world spatial tasks, despite the dataset’s synthetic and compact nature. Code and data are available at https://github.com/rsoohyun/SpatialBlock.

View arXiv pageView PDFGitHub0Add to collection

Get this paper in your agent:

hf papers read 2609\.07064

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper6

#### rsoohyun/SpatialBlock-3B-direct Image-Text-to-Text• 4B• Updated1 day ago • 39 #### rsoohyun/SpatialBlock-7B-direct Image-Text-to-Text• 8B• Updated1 day ago • 40 #### rsoohyun/SpatialBlock-4B-direct Image-Text-to-Text• 4B• Updated1 day ago • 35 #### rsoohyun/SpatialBlock-3B-reason Image-Text-to-Text• 4B• Updated1 day ago • 40 Browse 6 models citing this paper## Datasets citing this paper1

#### rsoohyun/SpatialBlock-15k Viewer• Updated1 day ago • 15.6k • 57 • 1

Spaces citing this paper1

Collections including this paper1

Similar Articles

OVO-S-Bench: A Hierarchical Benchmark for Streaming Spatial Intelligence in Multimodal LLMs

Papers with Code Trending

OVO-S-Bench introduces a comprehensive human-annotated benchmark of 1,680 questions across 348 videos to evaluate streaming spatial intelligence in multimodal LLMs, revealing that even the best model (Gemini-3.1-Pro) trails human experts by 27 points. The benchmark exposes key limitations including allocentric mapping as a major bottleneck and chain-of-thought reasoning amplifying spatial errors.