SyncAI.news, a Varaisys broadcasting
Groundbench: Multi-Resolution Polygon Grounding Exposes the Geometry Gap in Vision-Language Models
ZB

Zhonghan Bian, Zhenran Wang, Jinsong Li, Zhangyang Qi

· 1 min read

ResearcharXiv cs.CV

Groundbench: Multi-Resolution Polygon Grounding Exposes the Geometry Gap in Vision-Language Models

arXiv:2609.27821v1 Announce Type: new Abstract: Bounding-box scores on RefCOCO-family grounding leave little room to distinguish frontier vision-language systems, yet boxes discard object shape. We introduce GroundingBench, a matched benchmark that re-targets the same 1,500 image-expression-referent triples to exact-N polygons at five vertex budgets. A fixed-denominator harness separately audits filled-region intersection over union (IoU) and legal-polygon completion. The strongest tested configuration reaches 88.2 box IoU and 97.1 accuracy at IoU >= .5 (Acc@.5), versus 57.7 and 69.2 for direct polygons; because these headline scores use different references, we also compare direct polygons with predicted boxes rasterised against the same contour target, obtaining 57.7 versus 57.3 when pooled. Performance is non-monotone in N and collapses at the densest budget, where legality failures compound residual geometric error. Qwen's thinking-setting contrast is the largest tested input-preserving configuration difference; under frozen templates, false spatial cues are more damaging than false colour cues, and target preference can remain high while contour tracing is poor. Alternate masks and a continuous-area scorer preserve the principal ordering. GroundingBench therefore measures an operational output-geometry gap spanning localisation, boundary construction, serialisation, and topology, rather than latent boundary perception alone.

Original source

This story was published by arXiv cs.CV and written by Zhonghan Bian, Zhenran Wang, Jinsong Li, Zhangyang Qi. SyncAI.news shows a preview; the complete article is on the publisher's site.

Read the full story on arxiv.org

Similar News