SAILenv 3D Virtual Environments

该数据集包含了逼真的3D虚拟环境，其中每个像素都有语义标签和运动向量，这些环境可以无限生成，旨在评估我们提出的方法。为了衡量代理的能力，我们创建了三种不同场景：空旷空间、固体环境以及客厅，这些场景在复杂性和背景上各不相同。数据集规模涵盖三个3D场景，包含多个对象交互。任务是对每个帧像素进行开放集类别增量分类。

This dataset comprises photorealistic 3D virtual environments, where each pixel is annotated with semantic labels and motion vectors. These environments can be infinitely generated, and are designed to evaluate our proposed method. To assess the capabilities of the AI agent, we created three distinct scenarios: open spaces, solid environments, and living rooms, which differ in complexity and background context. The dataset encompasses three 3D scenarios with multiple object interactions. The task is to perform open-set incremental class classification on each pixel of every frame.