Uncertainty-Aware Gradient Stabilization for Small Object Detection
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2023
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866912473773244416 |
|---|---|
| author | Sun, Huixin Li, Yanjing Yang, Linlin Cao, Xianbin Zhang, Baochang |
| author_facet | Sun, Huixin Li, Yanjing Yang, Linlin Cao, Xianbin Zhang, Baochang |
| contents | Despite advances in generic object detection, there remains a performance gap in detecting small objects compared to normal-scale objects. We reveal that conventional object localization methods suffer from gradient instability in small objects due to sharper loss curvature, leading to a convergence challenge. To address the issue, we propose Uncertainty-Aware Gradient Stabilization (UGS), a framework that reformulates object localization as a classification task to stabilize gradients. UGS quantizes continuous labels into interval non-uniform discrete representations. Under a classification-based objective, the localization branch generates bounded and confidence-driven gradients, mitigating instability. Furthermore, UGS integrates an uncertainty minimization (UM) loss that reduces prediction variance and an uncertainty-guided refinement (UR) module that identifies and refines high-uncertainty regions via perturbations. Evaluated on four benchmarks, UGS consistently improves anchor-based, anchor-free, and leading small object detectors. Especially, UGS enhances DINO-5scale by 2.6 AP on VisDrone, surpassing previous state-of-the-art results. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2303_01803 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Uncertainty-Aware Gradient Stabilization for Small Object Detection Sun, Huixin Li, Yanjing Yang, Linlin Cao, Xianbin Zhang, Baochang Computer Vision and Pattern Recognition Despite advances in generic object detection, there remains a performance gap in detecting small objects compared to normal-scale objects. We reveal that conventional object localization methods suffer from gradient instability in small objects due to sharper loss curvature, leading to a convergence challenge. To address the issue, we propose Uncertainty-Aware Gradient Stabilization (UGS), a framework that reformulates object localization as a classification task to stabilize gradients. UGS quantizes continuous labels into interval non-uniform discrete representations. Under a classification-based objective, the localization branch generates bounded and confidence-driven gradients, mitigating instability. Furthermore, UGS integrates an uncertainty minimization (UM) loss that reduces prediction variance and an uncertainty-guided refinement (UR) module that identifies and refines high-uncertainty regions via perturbations. Evaluated on four benchmarks, UGS consistently improves anchor-based, anchor-free, and leading small object detectors. Especially, UGS enhances DINO-5scale by 2.6 AP on VisDrone, surpassing previous state-of-the-art results. |
| title | Uncertainty-Aware Gradient Stabilization for Small Object Detection |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2303.01803 |