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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add global map-consistency reward (ablation) | GRPO with answer reward · MindCube-Tiny | accuracy: 63.8 -> 64.9 | Improved |
| add local trajectory reward (ablation) | GRPO with answer+global reward · MindCube-Tiny | accuracy: 64.9 -> 66.5 | Improved |
| add GRPO with dense rewards after SFT | DR-MV3D · MindCube-Tiny | accuracy: 62.4 -> 66.5 | Improved |
| add GRPO with dense rewards after SFT | DR-MV3D · VSI-Bench | Avg: 34.2 -> 37.1 | Improved |
| add GRPO with dense rewards after SFT | DR-MV3D · BLINK (MV) | accuracy: 54.9 -> 56.4 | Improved |
| add trajectory supervision to both maps in SFT (ablation) | DR-MV3D SFT · MindCube-Tiny | accuracy: 53.6 -> 62.4 | Improved |
| supervise trajectory only vs maps only (ablation) | DR-MV3D SFT · MindCube-Tiny | accuracy: 52.8 -> 58.2 | Improved |
| supervise both maps instead of allocentric alone (ablation) | DR-MV3D SFT · MindCube-Tiny | accuracy: 52.8 -> 53.6 | No clear effect |
| train with GT annotations instead of VFM-only pseudo targets (ablation) | DR-MV3D SFT+GRPO · MindCube-Tiny | accuracy: 57.7 -> 66.5 | Improved |
| use DR-MV3D vs map-based GRPO baseline | DR-MV3D · MindCube-Tiny | accuracy: 53.7 -> 66.5 vs CGMap-FFR-RL | Improved |
| use DR-MV3D vs InternVL2-8B on App. Order | DR-MV3D · VSI-Bench | App. Order: 46.4 vs 31.4 | Hurt |
| use DR-MV3D vs GPT-5 on Rotation | DR-MV3D · MindCube-Tiny Rotation | accuracy: 94.5 vs 43.0 | Hurt |
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.
- RoBoBrain — RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete (CVPR 2025) · better on 1 of 1 shared result
- Spatial-MLLM — Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 2025) · better on 1 of 1 shared result
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
- Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA (AAAI 2024 · similarity 0.91)
- 3D Question Answering via only 2D Vision-Language Models (ICML 2025 · similarity 0.90)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.90)
- Hypo3D: Exploring Hypothetical Reasoning in 3D (ICML 2025 · similarity 0.89)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.89)
- Vid-LLM: A Compact Video-based 3D Multimodal LLM with Reconstruction–Reasoning Synergy (ICLR 2026 · similarity 0.89)
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.
