Trans2k

Trans2k是首个针对透明物体追踪的训练数据集，由卢布尔雅那大学计算机与信息科学学院创建。该数据集包含超过2000个序列和104,343张图像，均通过边界框和分割掩码进行标注。数据集内容丰富，涵盖多种透明物体和复杂场景，通过现代渲染引擎生成，确保了图像的真实性和多样性。创建过程中，研究者精心设计了多种视觉属性，如透明度、运动模糊和部分遮挡等，以模拟真实世界中的挑战。Trans2k的应用领域广泛，主要用于提升深度学习追踪架构在透明物体追踪任务中的性能，解决现有追踪器在透明物体上表现不佳的问题。

Trans2k is the first training dataset dedicated to transparent object tracking, created by the Faculty of Computer and Information Science, University of Ljubljana. This dataset contains over 2000 sequences and 104,343 images, all annotated with bounding boxes and segmentation masks. Featuring rich content covering diverse transparent objects and complex scenarios, the dataset is generated using modern rendering engines to ensure the authenticity and diversity of the images. During the creation process, researchers carefully designed various visual attributes such as transparency, motion blur, and partial occlusion to simulate real-world challenges. Trans2k has a wide range of application scenarios, mainly used to improve the performance of deep learning tracking architectures on transparent object tracking tasks, addressing the poor performance of existing trackers on transparent objects.