ScanNet200
收藏资源简介:
ScanNet200数据集包含了200个自然类别不平衡的3D场景。该数据集由穆罕默德·本·扎耶德人工智能大学创建,旨在为3D实例分割任务提供具有真实世界动态的增量学习场景。数据集涵盖了多种室内环境中的物体类别,如桌子、椅子、沙发、枕头等。通过设计基于类频率、语义相似性和随机分组的三个增量场景,以模拟新类别随时间连续出现并具有自然类别不平衡的实际世界情况。
The ScanNet200 dataset consists of 200 natural class-imbalanced 3D scenes. Developed by Mohammed bin Zayed University of Artificial Intelligence (MBZUAI), it is designed to provide real-world dynamic incremental learning scenarios for 3D instance segmentation tasks. The dataset covers a wide range of object categories in indoor environments, including tables, chairs, sofas, pillows and other common items. Three incremental scenarios are constructed based on class frequency, semantic similarity and random grouping, aiming to simulate real-world scenarios where new categories emerge continuously over time while exhibiting natural class imbalance.




