[论文解读] Select-Mosaic: Data Augmentation Method for Dense Small Object Scenes

Hao Zhang, Shuaijie Zhang+1 more|arXiv (Cornell University)|Jun 8, 2024

Computer Graphics and Visualization Techniques被引用 5

一句话总结

Select-Mosaic 是一种改进的 Mosaic 数据增强技术，具有细粒度区域选择策略，旨在提升对航空图像中密集分布的小目标的检测能力。

ABSTRACT

Data augmentation refers to the process of applying a series of transformations or expansions to original data to generate new samples, thereby increasing the diversity and quantity of the data, effectively improving the performance and robustness of models. As a common data augmentation method, Mosaic data augmentation technique stitches multiple images together to increase the diversity and complexity of training data, thereby reducing the risk of overfitting. Although Mosaic data augmentation achieves excellent results in general detection tasks by stitching images together, it still has certain limitations for specific detection tasks. This paper addresses the challenge of detecting a large number of densely distributed small objects in aerial images by proposing the Select-Mosaic data augmentation method, which is improved with a fine-grained region selection strategy. The improved Select-Mosaic method demonstrates superior performance in handling dense small object detection tasks, significantly enhancing the accuracy and stability of detection models. Code is available at https://github.com/malagoutou/Select-Mosaic.

研究动机与目标

• Motivate improved detection of densely distributed small objects in aerial imagery.
• Address limitations of standard Mosaic augmentation for dense scenes.
• Develop a fine-grained region selection strategy to enhance augmentation quality.
• Improve detection accuracy and stability for small object-rich scenes.

提出的方法

• Build on Mosaic data augmentation by introducing a fine-grained region selection strategy.
• Stitch multiple images similarly to Mosaic but select regions with higher relevance for small objects.
• Demonstrate improved performance on dense small object detection tasks.
• Provide code availability to enable reproducibility.

实验结果

研究问题

• RQ1 Can Select-Mosaic improve detection accuracy for densely distributed small objects compared to standard Mosaic?
• RQ2 How does the fine-grained region selection strategy affect model robustness and stability in dense small-object scenes?
• RQ3 Is Select-Mosaic effective for aerial imagery with numerous small objects?
• RQ4 What is the impact of Select-Mosaic on training diversity and overfitting risk?

主要发现

• Select-Mosaic demonstrates superior performance in handling dense small object detection tasks.
• The approach enhances accuracy and stability of detection models in dense small-object scenarios.
• The method improves training data diversity while focusing on relevant small-object regions.