AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model

Youngwan Lee, Soojin Jang, Yoorhim Cho, Seunghwan Lee, Yong-Ju Lee, Sung Ju Hwang

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
post-train with GRPO on MultihopSpatial-TrainQwen3-VL-4B-Instruct · MultihopSpatialMCQ accuracy (%): 37.8 -> 62.9Improved
post-train with GRPO on MultihopSpatial-TrainQwen3-VL-4B-Instruct · MultihopSpatialAcc@50IoU (%): 31.0 -> 53.8Improved
post-train with GRPO on MultihopSpatial-TrainQwen3-VL-4B-Instruct · BLINKaccuracy (%): 82.5 -> 85.3Improved
post-train with GRPO on MultihopSpatial-TrainQwen3-VL-4B-Instruct · 3DSRBenchaccuracy (%): 56.1 -> 56.3No clear effect
use MultihopSpatial-Train as VLA backbone trainingQwen3-VL-4B-Instruct · CALVIN ABC-Davg. completed tasks: 3.75 -> 3.98Improved
use MultihopSpatial-Train as VLA backbone trainingQwen3-VL-4B-Instruct · LIBERO-10success rate (%): 35.8 -> 40.0Improved
add GRPO with MCQ reward on top of SFT (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)MCQ accuracy (%): 58.1 -> 61.8Improved
add bounding box reward (GIoU) to GRPO (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)Acc@50IoU (%): 51.6 -> 52.3Improved
add bounding box reward (GIoU) to GRPO (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)MCQ accuracy (%): 61.8 -> 61.1Hurt
use GRPO with bbox reward instead of SFT (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)Acc@50IoU (%): 49.0 -> 52.3Improved
increase reward weight alpha from 1 to 2 (beta=1) (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)MCQ accuracy (%): 61.1 -> 52.5Hurt
increase reward weight beta from 1 to 2 (alpha=1) (ablation)Qwen3-VL-4B-Instruct · MultihopSpatial (5-epoch)Acc@50IoU (%): 52.3 -> 51.1Hurt

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