Saved in:
Bibliographic Details
Main Authors: Manoharan, Avinaash, Yin, Xiangyu, Helm, Domenik, Cheng, Chih-Hong
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.12871
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908872744108032
author Manoharan, Avinaash
Yin, Xiangyu
Helm, Domenik
Cheng, Chih-Hong
author_facet Manoharan, Avinaash
Yin, Xiangyu
Helm, Domenik
Cheng, Chih-Hong
contents Evaluating object detection models in deployment is challenging because ground-truth annotations are rarely available. We introduce the Cumulative Consensus Score (CCS), a label-free monitoring signal for continuous evaluation and comparison of detectors in real-world settings. CCS applies test-time data augmentation to each image and measures the spatial consistency of predicted bounding boxes across augmented views using Intersection over Union. The resulting consensus score serves as a proxy for reliability without requiring bounding box annotations. In controlled experiments on Open Images and KITTI, CCS achieved over 90% congruence with F1-score, Probabilistic Detection Quality, and Optimal Correction Cost, with qualitative consistency further confirmed on COCO and BDD100K across model pairs. The method is model-agnostic, working across single-stage and two-stage detectors, and operates at the case level to highlight under-performing scenarios. We also provide a simplified theoretical link between expected CCS and detection correctness. Altogether, CCS provides a robust foundation for DevOps-style monitoring of object detectors.
format Preprint
id arxiv_https___arxiv_org_abs_2509_12871
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cumulative Consensus Score: Label-Free and Model-Agnostic Evaluation of Object Detectors in Deployment
Manoharan, Avinaash
Yin, Xiangyu
Helm, Domenik
Cheng, Chih-Hong
Computer Vision and Pattern Recognition
Evaluating object detection models in deployment is challenging because ground-truth annotations are rarely available. We introduce the Cumulative Consensus Score (CCS), a label-free monitoring signal for continuous evaluation and comparison of detectors in real-world settings. CCS applies test-time data augmentation to each image and measures the spatial consistency of predicted bounding boxes across augmented views using Intersection over Union. The resulting consensus score serves as a proxy for reliability without requiring bounding box annotations. In controlled experiments on Open Images and KITTI, CCS achieved over 90% congruence with F1-score, Probabilistic Detection Quality, and Optimal Correction Cost, with qualitative consistency further confirmed on COCO and BDD100K across model pairs. The method is model-agnostic, working across single-stage and two-stage detectors, and operates at the case level to highlight under-performing scenarios. We also provide a simplified theoretical link between expected CCS and detection correctness. Altogether, CCS provides a robust foundation for DevOps-style monitoring of object detectors.
title Cumulative Consensus Score: Label-Free and Model-Agnostic Evaluation of Object Detectors in Deployment
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.12871