AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Why Far Looks Up: Probing Spatial Representation in Vision-Language Models

Cheolhong Min, Jaeyun Jung, Daeun Lee, Hyeonseong Jeon, Yu Su, Jonathan Tremblay, Chan Hee Song, Jaesik Park

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 4 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
evaluate counter-heuristic vs consistent depth examples (no change to model)Molmo, NVILA, Qwen2.5-VL, Qwen3-VL · EmbSpatial-Benchaccuracy: Qwen2.5-VL +2M: 60.9 -> 24.0 (consistent -> counter)Hurt
scale spatial fine-tuning data 80k -> 2M (ablation)NVILA-Lite-2B · EmbSpatial-Bench counteraccuracy: 15.5 -> 41.1Improved
scale spatial fine-tuning data 80k -> 2M (ablation)Qwen2.5-VL-3B · EmbSpatial-Bench counteraccuracy: 30.2 -> 24.0Hurt
scale spatial fine-tuning data 80k -> 2M (ablation)Molmo-7B · EmbSpatial-Bench counteraccuracy: 29.5 -> 39.5 (400k: 27.1)Mixed
scale spatial fine-tuning data to 2M (ablation)Qwen2.5-VL-3B · BLINK Spatial Relationaccuracy: 83.9 -> 78.3Hurt
spatial fine-tuning data scaling (ablation)Molmo, NVILA, Qwen · five spatial benchmarksaccuracy: NVILA 2M: 93.8 CV-3D Depth vs 62.9 BLINK Spat. Rel.Mixed
very large-scale training (RoboRefer 20M+ QA, Qwen3-VL-235B)VLMs · SpatialTunnelaccuracy gap: +0.046 and +0.068 gaps; Qwen3 mean 0.908Improved
enable reasoning (ablation)GPT-5.2 · SpatialTunnel (exact match)accuracy: 0.613 -> 0.953Improved
higher distance-axis coherenceVLM representations · EmbSpatial-Benchcounter accuracy: rho = 0.759 (EmbSpatial), 0.804 (CV-Bench-3D)Improved
fine-tuning on spatial data increases vertical axis coherence (ablation)Molmo / Qwen · EmbSpatial-Bench probingvertical coherence: Molmo 0.23 -> 0.57; Qwen 0.29 -> 0.59Improved
RoboRefer vs NVILA 2M (richer training, lower VD-EI)NVILA-Lite-2B family · EmbSpatial-Bench counteraccuracy: 41.1 -> 59.7Improved

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