AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Perspective-Aware Reasoning in Vision-Language Models via Mental Imagery Simulation

Phillip Y. Lee, Jihyeon Je, Chanho Park, Mikaela Angelina Uy, Leonidas Guibas, Minhyuk Sung

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 33 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
apply abstract perspective change with visual promptQwen2.5-VL · COMFORT++ left/rightaccuracy (%): 43.33 -> 89.67Improved
apply abstract perspective change with numerical promptQwen2.5-VL · COMFORT++ left/rightaccuracy (%): 43.33 -> 88.67Improved
apply abstract perspective change with visual promptQwen2.5-VL · COMFORT++ facingaccuracy (%): 43.00 -> 88.33Improved
apply abstract perspective change with visual promptQwen2.5-VL · 3DSRBench left/rightaccuracy (%): 34.96 -> 72.78Improved
apply abstract perspective change with visual promptQwen2.5-VL · 3DSRBench visibilityaccuracy (%): 45.06 -> 67.44Improved
apply abstract perspective change with visual promptQwen2.5-VL · 3DSRBench facingaccuracy (%): 53.47 -> 66.47Improved
use visual prompt instead of numerical promptAPC · COMFORT++ visibilityaccuracy (%): 71.25 -> 90.00Improved
use visual prompt instead of numerical promptAPC · COMFORT++ facingaccuracy (%): 62.00 -> 88.33Improved
use numerical prompt instead of visual promptAPC · COMFORT++ closeraccuracy (%): 94.33 -> 96.00Hurt
use abstraction-based APC instead of dense reconstruction ViewCrafterViewCrafter · 3DSRBench left/rightaccuracy (%): 28.41 -> 72.78Improved
use APC abstraction instead of SpatialPIN* dense reconstructionSpatialPIN* · COMFORT++ left/rightaccuracy (%): 59.80 -> 89.67Improved
use APC instead of dense reconstruction in inference timeSpatialPIN* · Inference time per questiontime (s): 336.21 -> 17.47Improved

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