DeformingThings4D

该数据集包含多种对象类别、形状和动作的序列，用于训练和测试运动场的预测。在训练过程中，随机选择了25%的网格顶点，而其余的则用于测试。该数据集规模为7个序列，每个序列包含100个连续帧。任务是基于已观测到的点轨迹来预测未见点的运动。This dataset comprises sequences covering diverse object categories, shapes and motion patterns, dedicated to training and testing motion field prediction tasks. During the training phase, 25% of the grid vertices are randomly selected, while the remaining vertices are reserved for testing. The dataset includes 7 sequences in total, with each sequence consisting of 100 consecutive frames. The task is to predict the motions of unseen points based on the observed point trajectories.