Search papers, labs, and topics across Lattice.
The paper identifies systematic flaws in current VLM spatial intelligence benchmarks due to reliance on point-cloud annotations and full-scene access assumptions, leading to incorrect or unanswerable question-answer pairs. To address this, they introduce ReVSI, a new benchmark with re-annotated objects and geometry across 381 scenes, along with regenerated QA pairs verified by humans. Evaluations using ReVSI reveal previously obscured failure modes in VLMs, providing a more accurate assessment of spatial reasoning.
Current VLM spatial reasoning benchmarks are misleading, as they often penalize models for "incorrect" answers that are actually correct given the limited visual information the models receive.
Current evaluations of spatial intelligence can be systematically invalid under modern vision-language model (VLM) settings. First, many benchmarks derive question-answer (QA) pairs from point-cloud-based 3D annotations originally curated for traditional 3D perception. When such annotations are treated as ground truth for video-based evaluation, reconstruction and annotation artifacts can miss objects that are clearly visible in the video, mislabel object identities, or corrupt geometry-dependent answers (e.g., size), yielding incorrect or ambiguous QA pairs. Second, evaluations often assume full-scene access, while many VLMs operate on sparsely sampled frames (e.g., 16-64), making many questions effectively unanswerable under the actual model inputs. We improve evaluation validity by introducing ReVSI, a benchmark and protocol that ensures each QA pair is answerable and correct under the model's actual inputs. To this end, we re-annotate objects and geometry across 381 scenes from 5 datasets to improve data quality, and regenerate all QA pairs with rigorous bias mitigation and human verification using professional 3D annotation tools. We further enhance evaluation controllability by providing variants across multiple frame budgets (16/32/64/all) and fine-grained object visibility metadata, enabling controlled diagnostic analyses. Evaluations of general and domain-specific VLMs on ReVSI reveal systematic failure modes that are obscured by prior benchmarks, yielding a more reliable and diagnostic assessment of spatial intelligence.
