Balance18
收藏资源简介:
Balance18是由中国海洋大学信息科学与工程学院创建的一个针对水下物体检测的平衡数据集。该数据集通过类智能风格增强(CWSA)算法从URPC2018数据集中生成,旨在解决水下物体检测中的类别不平衡问题。数据集包含2897张图像,涵盖了海参、海胆、扇贝和海星四个类别,每个类别的样本数量经过平衡处理。创建过程中,使用了CycleGAN进行风格转换,以增加图像的多样性。Balance18数据集的应用领域主要集中在水下物体识别和检测,通过提供平衡的训练数据,有助于提高深度神经网络在实际水下环境中的泛化能力。
Balance18 is a balanced dataset tailored for underwater object detection, developed by the College of Information Science and Engineering, Ocean University of China. It is generated from the URPC2018 dataset via the Class-wise Smart Style Augmentation (CWSA) algorithm, with the core goal of addressing the class imbalance problem in underwater object detection. The dataset contains 2897 images covering four categories: sea cucumber, sea urchin, scallop, and starfish, with the sample size of each category being strictly balanced. During its construction, CycleGAN was utilized for style transfer to augment the diversity of the included images. The primary application scenarios of the Balance18 dataset focus on underwater object recognition and detection. By providing balanced training data, it facilitates the improvement of the generalization performance of deep neural networks in real-world underwater environments.




