TPA-Net
收藏资源简介:
TPA-Net数据集由加州大学洛杉矶分校创建,旨在通过高分辨率的3D物理模拟,包括固体和流体的模拟,以及文本描述物理现象,推动文本到视频/模拟(T2V/S)技术的发展。数据集利用了先进的物理模拟方法,如增量势接触(IPC)和材料点方法(MPM),模拟了多种物理现象,如弹性变形、材料断裂、碰撞和湍流等。此外,还提供了高质量的多视角渲染视频,以支持T2V、神经辐射场(NeRF)等研究。TPA-Net是自动化文本到视频/模拟(T2V/S)技术的第一步,旨在解决现有数据集在物理现实性方面的不足,为多模态生成研究提供高质量数据支持。
The TPA-Net dataset was developed by the University of California, Los Angeles (UCLA). It aims to advance the development of text-to-video/simulation (T2V/S) technologies through high-resolution 3D physical simulations, including simulations of solids and fluids paired with textual descriptions of physical phenomena. The dataset leverages advanced physical simulation methods such as Incremental Potential Contact (IPC) and Material Point Method (MPM) to simulate a diverse range of physical phenomena, including elastic deformation, material fracture, collisions, and turbulence. Additionally, high-quality multi-view rendered videos are provided to support research in fields like T2V and Neural Radiance Fields (NeRF). TPA-Net represents the first step toward automated text-to-video/simulation (T2V/S) technologies, designed to address the limitations of existing datasets in terms of physical realism, and provides high-quality data support for multimodal generation research.




