Research map / Vision-language and multimodal / Multimodal LLMs
Attention-based Vision-Language Memory for Spatial Reasoning
Zuntao Liu, Zuntao Liu, Taimeng Fu, Shaoshu Su, Cherie Ho, Chen Wang
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 3D-aware representation (ablation) | LLaVA-NeXT-Video-7B baseline · VSI-Bench avg | accuracy: 55.2 -> 63.8 | Improved |
| add working memory (ablation) | VLM2 3D-aware · VSI-Bench avg | accuracy: 63.8 -> 65.9 | Improved |
| add episodic memory (ablation) | VLM2 3D-aware · VSI-Bench avg | accuracy: 63.8 -> 66.1 | Improved |
| add both memory modules (ablation) | VLM2 3D-aware · VSI-Bench avg | accuracy: 63.8 -> 67.8 | Improved |
| use cross-attention instead of concat-MLP fusion (ablation) | pi3 geometry · VSI-Bench avg | accuracy: 40.4 -> 61.0 | Improved |
| add adaptive gate to 3D position injection (ablation) | A3PI · VSI-Bench avg | accuracy: 58.9 -> 61.6 | Improved |
| add viewpoint-aware geometry alignment (ablation) | Cross-Attn pi3 baseline · VSI-Bench avg | accuracy: 61.0 -> 62.9 | Improved |
| combine A3PI and VAGA (ablation) | Cross-Attn pi3 baseline · VSI-Bench avg | accuracy: 61.0 -> 63.8 | Improved |
| use pi3 instead of VGGT geometry with cross-attention (ablation) | Cross-Attn fusion · VSI-Bench avg | accuracy: 60.2 -> 61.0 | Improved |
| increase episodic memory capacity from 8 to 32 (ablation) | VLM2 (Lw=8) · VSI-Bench avg | accuracy: 66.3 -> 68.8 | Improved |
| increase working memory window from 8 to 16 (ablation) | VLM2 (Le=32) · VSI-Bench avg | accuracy: 68.8 -> 68.3 | Hurt |
| use VLM2 instead of VLM-3R on long videos | VSI-Bench · VSI-Bench long | accuracy: 60.0 -> 68.1 | Improved |
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.
- VLM-3R-7B — VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026) · better on 4 of 4 shared results
- Spatial-MLLM-4B — Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 2025) · better on 2 of 2 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
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.93)
- Spatial Understanding from Videos: Structured Prompts Meet Simulation Data (NeurIPS 2025 · similarity 0.92)
- Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 2025 · similarity 0.92)
- Is A Picture Worth A Thousand Words? Delving Into Spatial Reasoning for Vision Language Models (NeurIPS 2024 · similarity 0.92)
- Linear Mechanisms for Spatiotemporal Reasoning in Vision Language Models (ICLR 2026 · similarity 0.91)
- SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.91)
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.
