AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

SITE: towards Spatial Intelligence Thorough Evaluation

Wenqi Wang, Reuben Tan, Pengyue Zhu, Jianwei Yang, Zhengyuan Yang, Lijuan Wang, Andrey Kolobov, Jianfeng Gao, Boqing Gong

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Spatial reasoning in VLMs · 18 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
scale LLAVA-OneVision from 0.5B to 7BLLAVA-OneVision · SITE fullchance-adjusted accuracy: 18.4 -> 30.2Improved
scale Qwen2.5-VL from 3B to 7BQwen2.5-VL · SITE fullchance-adjusted accuracy: 29.5 -> 31.4Improved
scale InternVL-2.5 from 4B to 8BInternVL-2.5 · SITE fullchance-adjusted accuracy: 29.4 -> 32.8Improved
compare small open-source Qwen2.5-VL-7B and InternVL-2.5-8B with proprietary GPT-4oopen-source VLMs · SITE fullchance-adjusted accuracy: 31.4 and 32.8 vs 37.8No clear effect
use ego-exo view association instead of standard tasksGPT-4o · SITE view associationchance-adjusted accuracy: 35.70 and 20.70 vs human 100Hurt
use frames reordering instead of view associationGPT-4o · SITE new taskschance-adjusted accuracy: 35.70 -> -2.01 (ego2exo)Hurt
compare Qwen2.5-VL-7B with GPT-4o on frames reorderingQwen2.5-VL-7B · SITE frames reorderingchance-adjusted accuracy: 7.63 vs -2.01 (ego2exo)Improved
evaluate multi-view reasoning on full splitstate-of-the-art VLMs · SITE Multi-Viewchance-adjusted accuracy: 22.0 (GPT-4o) vs 87.5 (human, tiny)Hurt
fine-tune Qwen2.5-VL-7B instead of smaller VLMs on LIBERO-SpatialVLM backbone · LIBERO-Spatialsuccess rate (%): 0.0 -> 38.0Improved
correlate SITE score with robot manipulation performanceSITE · LIBERO-SpatialPearson correlation: 0.902Improved
correlate POPE score with robot manipulation performancePOPE · LIBERO-SpatialPearson correlation: -0.602Hurt
correlate MathVista score with robot manipulation performanceMathVista · LIBERO-SpatialPearson correlation: 0.935Improved

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