AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

MindBlock: Probing Spatial Assembly and Structure in Unified Multimodal Models

Baiqiao Yin, Junhao Liu, Han Yin, Heyang Yu, Tingxuan Zhang, Zhiheng Li, Chengzu Li, Jihan YANG, Manling Li, Chen Feng, Yiming Li

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add textual chain-of-thought before image generation (UniCoT vs BAGEL direct)BAGEL · MindBlock single-step assemblyChamfer distance: 12.4 -> 6.5 (-47.6%)Improved
add textual chain-of-thought (UniCoT vs BAGEL direct)BAGEL · MindBlock single-step assembly3DGS PSNR: 13.42 -> 20.88Improved
add textual chain-of-thought (UniCoT vs BAGEL direct)BAGEL · MindBlock single-step assemblyCLIP-S: 0.782 -> 0.814Improved
add textual chain-of-thought (UniCoT vs BAGEL direct)BAGEL · MindBlock single-step assembly3DGS SSIM: 0.712 -> 0.785Improved
switch from Spatial Assembly to Spatial Structure (viewpoint transformation)BAGEL · MindBlock3DGS PSNR: 13.83 -> 12.26Hurt
switch from Spatial Assembly to Spatial Structure (viewpoint transformation)Step1X-Edit · MindBlock3DGS PSNR: 15.20 -> 13.80Hurt
fine-tune with MindBlock-Semantic curriculum instead of baselineBAGEL-zebra-CoT · MindBlock forward assemblyChamfer distance: 11.8 -> 5.9Improved
fine-tune with MindBlock-Semantic curriculum instead of baselineBAGEL-zebra-CoT · MindBlock reverse deconstructionChamfer distance: 13.2 -> 6.8Improved
use semantic structures instead of random structures for trainingMindBlock fine-tuning · MindBlock forward assemblyPSNR: 17.8 -> 20.4Improved
go from perceiving to executing the next stepBAGEL · MindBlock single-step assemblyaccuracy: 52.5% -> 28.2%Hurt
rely on 2D CLIP-S ranking instead of 3DGS metricsBlip3-o-next vs BAGEL · MindBlock assemblyCLIP-S vs Chamfer: CLIP-S 0.81 vs 0.77; CD 13.05 vs 11.72Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy