References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: Achieving Domain Generalization in Underwater Object Detection by Image Stylization and Domain Mixup
(Submitted on 6 Apr 2021 (this version), latest version 18 Jan 2023 (v6))
Abstract: The performance of existing underwater object detection methods degrades seriously when facing domain shift problem caused by complicated underwater environments. Due to the limitation of the number of domains in the dataset, deep detectors easily just memorize a few seen domain, which leads to low generalization ability. Ulteriorly, it can be inferred that the detector trained on as many domains as possible is domain-invariant. Based on this viewpoint, we propose a domain generalization method from the aspect of data augmentation. First, the style transfer model transforms images from one source domain to another, enriching the domain diversity of the training data. Second, interpolating different domains on feature level, new domains can be sampled on the domain manifold. With our method, detectors will be robust to domain shift. Comprehensive experiments on S-UODAC2020 datasets demonstrate that the proposed method is able to learn domain-invariant representations, and outperforms other domain generalization methods. The source code is available at this https URL
Submission history
From: Pinhao Song [view email][v1] Tue, 6 Apr 2021 01:45:07 GMT (1421kb,D)
[v2] Sun, 9 May 2021 13:20:51 GMT (0kb,I)
[v3] Thu, 19 May 2022 02:07:31 GMT (2664kb,D)
[v4] Sun, 22 May 2022 05:49:53 GMT (2663kb,D)
[v5] Sat, 1 Oct 2022 03:54:26 GMT (3319kb)
[v6] Wed, 18 Jan 2023 09:24:01 GMT (11839kb)
Link back to: arXiv, form interface, contact.