Simulated Flying Shapes 和 Simulated Planar Manipulator

本研究发布了两个模拟数据集：Simulated Flying Shapes 和 Simulated Planar Manipulator，旨在评估视频处理系统的学习能力。Simulated Flying Shapes包含90000个灰度视频，展示两个相同形状和大小的物体，其中一个物体向另一个移动。Simulated Planar Manipulator则展示一个3自由度的平面机械臂执行拾取和放置任务。这两个数据集均涉及目标导向任务，不同于随机运动，更适合测试预测能力和机器学习模型的复杂运动学习。数据集创建过程中，物体的初始配置如位置、大小和颜色是随机采样的。这些数据集适用于深度神经网络模型的编码、重建或预测视频帧序列的测试，旨在解决机器学习模型在视频处理中的预测和学习问题。

This study introduces two simulated datasets: Simulated Flying Shapes and Simulated Planar Manipulator, which are intended to evaluate the learning capabilities of video processing systems. Simulated Flying Shapes comprises 90,000 grayscale videos showing two objects with identical shape and size, where one object moves toward the other. Simulated Planar Manipulator depicts a 3-degree-of-freedom planar robotic arm performing pick-and-place tasks. Both datasets involve goal-directed tasks, which differ from random motions and are more suitable for testing predictive abilities and complex motion learning of machine learning models. During the dataset generation process, the initial configurations of the objects, including position, size and color, are randomly sampled. These datasets are applicable for testing deep neural network models in encoding, reconstructing or predicting video frame sequences, and aim to solve the prediction and learning problems of machine learning models in video processing.