AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Dense Reward for Multi-View 3D Reasoning with Global Maps and Local Views

Jiho Choi, Seonho Lee, Seojeong Park, Hyunjung Shim

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 2 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 global map-consistency reward (ablation)GRPO with answer reward · MindCube-Tinyaccuracy: 63.8 -> 64.9Improved
add local trajectory reward (ablation)GRPO with answer+global reward · MindCube-Tinyaccuracy: 64.9 -> 66.5Improved
add GRPO with dense rewards after SFTDR-MV3D · MindCube-Tinyaccuracy: 62.4 -> 66.5Improved
add GRPO with dense rewards after SFTDR-MV3D · VSI-BenchAvg: 34.2 -> 37.1Improved
add GRPO with dense rewards after SFTDR-MV3D · BLINK (MV)accuracy: 54.9 -> 56.4Improved
add trajectory supervision to both maps in SFT (ablation)DR-MV3D SFT · MindCube-Tinyaccuracy: 53.6 -> 62.4Improved
supervise trajectory only vs maps only (ablation)DR-MV3D SFT · MindCube-Tinyaccuracy: 52.8 -> 58.2Improved
supervise both maps instead of allocentric alone (ablation)DR-MV3D SFT · MindCube-Tinyaccuracy: 52.8 -> 53.6No clear effect
train with GT annotations instead of VFM-only pseudo targets (ablation)DR-MV3D SFT+GRPO · MindCube-Tinyaccuracy: 57.7 -> 66.5Improved
use DR-MV3D vs map-based GRPO baselineDR-MV3D · MindCube-Tinyaccuracy: 53.7 -> 66.5 vs CGMap-FFR-RLImproved
use DR-MV3D vs InternVL2-8B on App. OrderDR-MV3D · VSI-BenchApp. Order: 46.4 vs 31.4Hurt
use DR-MV3D vs GPT-5 on RotationDR-MV3D · MindCube-Tiny Rotationaccuracy: 94.5 vs 43.0Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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