On Calibrating Deep Object Detectors

Date

2024-1-16

Author

Güngör, Muhammed Ertuğrul

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

150
views

53
downloads

Cite This

Recent years have seen unprecedented progress in Object Detection models. However, these advancements are typically limited to relatively balanced datasets. On long-tailed datasets, detectors often exhibit a bias towards head classes, resulting in subpar performance for tail classes. Long-tailed learning is crucial as the objects in real life follow this type of distribution. Numerous techniques have been proposed to address this issue. In this thesis, we review methods from the most influential branches of long-tailed learning. We then propose two post-hoc class score calibration methods that utilize training performance measurements, offering an alternative to existing methods that rely on class sample sizes. These methods update the probabilities or logits, using factors computed from different performance evaluation results. Furthermore, we introduce a third method that employs a ranking-based loss function during the second stage of training. We evaluate these methods using a challenging long-tailed dataset LVIS and compare our results with recent approaches. Our results demonstrate that our methods improve upon the baseline established with LVIS and present competitive performance compared to similar approaches.

Subject Keywords

Object Detection, Long-tailed Learning, Class Score Calibration, Logit Adjustment

URI

https://hdl.handle.net/11511/107826

Collections

Graduate School of Natural and Applied Sciences, Thesis