Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2604.15376 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914480420552704 |
|---|---|
| author | Kim, Keon Chelikavada, Krish |
| author_facet | Kim, Keon Chelikavada, Krish |
| contents | Multi-step zoom-in pipelines are widely used for GUI grounding, yet the intermediate predictions they produce are typically discarded after coordinate remapping. We observe that these intermediate outputs contain a useful confidence signal for free: zoom consistency, the distance between a model's step-2 prediction and the crop center. Unlike log-probabilities or token-level uncertainty, zoom consistency is a geometric quantity in a shared coordinate space, making it directly comparable across architecturally different VLMs without calibration. We prove this quantity is a linear estimator of step-1 spatial error under idealized conditions (perfect step-2, target within crop) and show it correlates with prediction correctness across two VLMs (AUC = 0.60; Spearman rho = -0.14, p < 10^{-6} for KV-Ground-8B; rho = -0.11, p = 0.0003 for Qwen3.5-27B). The correlation is small but consistent across models, application categories, and operating systems. As a proof-of-concept, we use zoom consistency to route between a specialist and generalist model, capturing 16.5% of the oracle headroom between them (+0.8%, McNemar p = 0.19). Code is available at https://github.com/omxyz/zoom-consistency-routing. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_15376 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Zoom Consistency: A Free Confidence Signal in Multi-Step Visual Grounding Pipelines Kim, Keon Chelikavada, Krish Computer Vision and Pattern Recognition Artificial Intelligence Multi-step zoom-in pipelines are widely used for GUI grounding, yet the intermediate predictions they produce are typically discarded after coordinate remapping. We observe that these intermediate outputs contain a useful confidence signal for free: zoom consistency, the distance between a model's step-2 prediction and the crop center. Unlike log-probabilities or token-level uncertainty, zoom consistency is a geometric quantity in a shared coordinate space, making it directly comparable across architecturally different VLMs without calibration. We prove this quantity is a linear estimator of step-1 spatial error under idealized conditions (perfect step-2, target within crop) and show it correlates with prediction correctness across two VLMs (AUC = 0.60; Spearman rho = -0.14, p < 10^{-6} for KV-Ground-8B; rho = -0.11, p = 0.0003 for Qwen3.5-27B). The correlation is small but consistent across models, application categories, and operating systems. As a proof-of-concept, we use zoom consistency to route between a specialist and generalist model, capturing 16.5% of the oracle headroom between them (+0.8%, McNemar p = 0.19). Code is available at https://github.com/omxyz/zoom-consistency-routing. |
| title | Zoom Consistency: A Free Confidence Signal in Multi-Step Visual Grounding Pipelines |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2604.15376 |