# Confidence Scores in Open-Vocabulary Detection Are a Biased Mixture of Scale and Semantics

arXiv: https://arxiv.org/abs/2607.10993 · HTML: https://mlsanity.com/paper/2607.10993/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 28 of its 34 references that are in the map.
- First posted on arXiv 13 Jul 2026.

## Builds on (most cited first, 28 of 28)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks](https://mlsanity.com/paper/1506.01497/) (2015)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [Feature Pyramid Networks for Object Detection](https://mlsanity.com/paper/1612.03144/) (2016)
- [On Calibration of Modern Neural Networks](https://mlsanity.com/paper/1706.04599/) (2017)
- [You Only Look Once: Unified, Real-Time Object Detection](https://mlsanity.com/paper/1506.02640/) (2015)
- [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://mlsanity.com/paper/2102.05918/) (2021)
- [Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://mlsanity.com/paper/2303.05499/) (2023)
- [SSD: Single Shot MultiBox Detector](https://mlsanity.com/paper/1512.02325/) (2015)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift](https://mlsanity.com/paper/1906.02530/) (2019)
- [LVIS: A Dataset for Large Vocabulary Instance Segmentation](https://mlsanity.com/paper/1908.03195/) (2019)
- [Open-vocabulary Object Detection via Vision and Language Knowledge Distillation](https://mlsanity.com/paper/2104.13921/) (2021)
- [Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMs](https://mlsanity.com/paper/2401.06209/) (2024)
- [FLAVA: A Foundational Language And Vision Alignment Model](https://mlsanity.com/paper/2112.04482/) (2021)
- [RegionCLIP: Region-based Language-Image Pretraining](https://mlsanity.com/paper/2112.09106/) (2021)
- [When and why vision-language models behave like bags-of-words, and what to do about it?](https://mlsanity.com/paper/2210.01936/) (2022)
- [YOLO-World: Real-Time Open-Vocabulary Object Detection](https://mlsanity.com/paper/2401.17270/) (2024)
- [Measuring Calibration in Deep Learning](https://mlsanity.com/paper/1904.01685/) (2019)
- [Open-Vocabulary Object Detection Using Captions](https://mlsanity.com/paper/2011.10678/) (2020)
- [Supervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm](https://mlsanity.com/paper/2110.05208/) (2021)
- [Scaling Open-Vocabulary Object Detection](https://mlsanity.com/paper/2306.09683/) (2023)
- [Bridging the Gap between Object and Image-level Representations for Open-Vocabulary Detection](https://mlsanity.com/paper/2207.03482/) (2022)
- [You Only Look One-level Feature](https://mlsanity.com/paper/2103.09460/) (2021)
- [Multivariate Confidence Calibration for Object Detection](https://mlsanity.com/paper/2004.13546/) (2020)
- [Training-free Boost for Open-Vocabulary Object Detection with Confidence Aggregation](https://mlsanity.com/paper/2404.08603/) (2024)

## Abstract (arXiv)

Foundation models such as CLIP have enabled open-vocabulary object detectors that generalise to novel categories via vision-language similarity. However, the confidence scores these detectors produce are not reliable localization probability estimates: they conflate visual scale and semantic query specificity with the true detection signal. Through controlled experiments on COCO across three foundation-model-based detectors (GroundingDINO, OWL-ViT, YOLO-World), with the scale-bias finding further replicated on LVIS (1,203 categories) using GroundingDINO, we show that s=cos(v,t) is a biased mixture of two effects. Scale bias (alpha = +0.064, r = 0.579, p = 1.29 x 10^-58) systematically inflates scores for large objects. Semantic bias (beta = -0.705, p = 5.23 x 10^-41) suppresses scores for generic queries. Both biases are structurally inevitable from CLIP's image-level pretraining. Threshold adjustment cannot remove them: oracle per-scale thresholding yields Delta F1 = +0.001 for small objects versus +0.102 for large. A parameter-free temperature scaling correction improves small-object Recall@10 by 19.6% (p < 0.01) without retraining. This comes at a modest, measurable cost to pooled-ranking precision, so the bias is partially, not freely, reversible at inference time. These findings reveal a fundamental limitation of adapting image-level foundation models to region-level detection tasks.