SSMNBench shows that MLLMs suffer distraction degradation on single-view-sufficient tasks and fail to integrate geometric evidence across views, instead relying on semantic averaging and view preference.
Why do mllms struggle with spatial understanding? a systematic analysis from data to architecture
6 Pith papers cite this work. Polarity classification is still indexing.
6
Pith papers citing it
years
2026 6representative citing papers
VLMs exhibit consistent vertical-distance entanglement in embeddings from perspective bias in natural images, producing accuracy gaps that a new synthetic benchmark SpatialTunnel exposes as model-intrinsic.
Orientation information is recoverable from MLLM visual encoder embeddings via linear regression, contradicting the hypothesis that failures originate in the encoders.
SCP defines a new benchmark task for predicting spatial causal outcomes beyond direct observation and shows that 23 leading models lag far behind humans on it.
SVoT uses RL with GRPO to train MLLMs on interleaved textual and visual reasoning chains for multi-hop spatial tasks, achieving up to 65% accuracy gains on new domains with quantitative state verification.
MPDocBench-Parse provides 433 annotated multi-page documents and an evaluation protocol covering text/table/formula extraction, merging, figure extraction, reading order, and heading hierarchy for realistic document parsing.
citing papers explorer
-
SSMNBench: Diagnosing Image-based Cross-View Human-Object Understanding via Single-View Sufficiency and Multi-View Necessity
SSMNBench shows that MLLMs suffer distraction degradation on single-view-sufficient tasks and fail to integrate geometric evidence across views, instead relying on semantic averaging and view preference.
-
Why Far Looks Up: Probing Spatial Representation in Vision-Language Models
VLMs exhibit consistent vertical-distance entanglement in embeddings from perspective bias in natural images, producing accuracy gaps that a new synthetic benchmark SpatialTunnel exposes as model-intrinsic.
-
Why MLLMs Struggle to Determine Object Orientations
Orientation information is recoverable from MLLM visual encoder embeddings via linear regression, contradicting the hypothesis that failures originate in the encoders.
-
SCP: Spatial Causal Prediction in Video
SCP defines a new benchmark task for predicting spatial causal outcomes beyond direct observation and shows that 23 leading models lag far behind humans on it.
-
SVoT: State-aware Visualization-of-Thought for Spatial Reasoning via Reinforcement Learning
SVoT uses RL with GRPO to train MLLMs on interleaved textual and visual reasoning chains for multi-hop spatial tasks, achieving up to 65% accuracy gains on new domains with quantitative state verification.
-
MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing
MPDocBench-Parse provides 433 annotated multi-page documents and an evaluation protocol covering text/table/formula extraction, merging, figure extraction, reading order, and heading hierarchy for realistic document parsing.