数据集:Dynamic Furry Animal Dataset

cat ├── img │ └── run - Motion name. │ └── %d - The frame number, start from 0. │ └──img_%04d.jpg - RGB images for each view. view number start from 0. │ └──img_%04d_alpha.png - Alpha mattes for corresponding RGB image. │ └── ... │ ├── volumes │ └── coords_init.pth - voxel coordinates represent an animal in rest pose. │ └── volume_indices.pth - The indices of bones to which the voxels are bound. | └── volume_weights.pth - The skinning weights of the voxels. | └── radius.txt - The radius of the volume. | ├── bones │ └── run - Motion name. │ └── Bones_%04d.inf - The skeletal pose for each frame, start from 0. In each row, the 3x4 [R T] matrix is displayed in columns, with the third column followed by columns 1, 2, and 4. │ └── ... | └── bones_parents.npy - The parent bone index of each bone. | └── Bones_0000.inf - The rest pose. | ├── CamPose.inf - Camera extrinsics. In each row, the 3x4 [R T] matrix is displayed in columns, with the third column followed by columns 1, 2, and 4, where R*X^{camera}+T=X^{world}. │ └── Intrinsic.inf - Camera intrinsics. The format of each intrinsics is: "idx \n fx 0 cx \n 0 fy cy \n 0 0 1 \n \n" (idx starts from 0) │ └── sequences - Motion sequences. The format of each motion is: "motion_name frames\n"

原始数据

艺术家建模了九种高质量的 CGI 动物,包括熊猫、狮子、猫等。
它们具有基于纤维/线的毛皮和骨骼。

数据集没有找到原始的CGI mesh和skinning

合成数据

使用商业渲染引擎(例如 MAYA)将所有这些 CGI 动物角色渲染成各种代表性骨骼运动下的高质量多视图 1080 × 1080 RGBA 视频。具体来说,我们采用了 36 个摄像机视图,这些摄像机视图均匀地围绕捕获的动物排列成一个圆圈,每个动物的代表性姿势数量从 700 到 1000 个不等。

人工标注

生成标注

  1. 基于Artemis论文方法生成体素以及体素的skinning weight
  2. 动作数据

Reference

  1. 论文:https://caterpillarstudygroup.github.io/ReadPapers/32.html
  2. 数据集下载地址:https://shanghaitecheducn-my.sharepoint.com/:f:/g/personal/luohm_shanghaitech_edu_cn/Et60lJpJdp5DoyQF7uzP6jgB_JEW4LIHixAyXEiVhHT3Vw?e=d09jtz
  3. github:https://github.com/HaiminLuo/Artemis