AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Reasoning Path and Latent State Analysis for Multi-view Visual Spatial Reasoning: A Cognitive Science Perspective

Qiyao Xue, Haoming Wang, Weichen Liu, Shiqi Wang, Yuyang Wu, Wei Gao

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 5 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
use object-centric instead of view-centric viewpoint patternVLMs · ReMindView-BenchAvg accuracy: 40.6 -> 37.0 (GPT-4o)Improved
scale Qwen2.5-VL from 3B to 32BQwen2.5-VL · ReMindView-BenchAvg accuracy: 32.2 -> 37.5Improved
scale InternVL3.5 from 8B to 38BInternVL3.5 · ReMindView-BenchAvg accuracy: 25.3 -> 32.6Improved
scale InternVL3.5 from 2B to 8BInternVL3.5 · ReMindView-BenchAvg accuracy: 31.6 -> 25.3Hurt
use spatial-specialized models vs general open-sourceSpatialVLM / SpaceQwen2.5-VL · ReMindView-BenchAvg accuracy: 31.6 and 28.4 vs 33.8 (LLaVA-OneVision-7B)No clear effect
require cross-frame reasoning vs within-frameVLMs · ReMindView-BenchaccuracyHurt
perspective-changing queries vs self-perspectiveVLMs · ReMindView-Benchaccuracy: drop >4%Hurt
increase number of objects in sceneVLMs · ReMindView-BenchaccuracyHurt
progress through reasoning phases (phase 1 to 3)Qwen2.5-VL-32B · ReMindView-Benchjudge correctness (%): 81.9 -> 38.8Hurt
scale Qwen2.5-VL from 3B to 32B (self-consistency)Qwen2.5-VL · ReMindView-Benchself-consistency overall (%): 48.6 -> 68.4Improved
correct vs incorrect answers on self-consistencyQwen2.5-VL-32B · ReMindView-Benchself-consistency (%): 54.2 -> 83.6Improved
increase model size on linear probe lossQwen2.5-VL family · ReMindView-Benchprobe cross-entropyHurt

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