SAM 3D Objects 多视角微调(reconviagensam3d 分支)所用的训练与评估数据。 每个卷是一个 tar.zst,内部路径 = 原始绝对路径去掉开头的 /。解压到 / 即把 数据放回与原机完全一致的位置,所有符号链接自动生效,无需 relink: 或用仓库里的 restore.sh(支持按分组还原、边解边删控制磁盘占用)。 注意第二行:物体的 images/ 链接走的是 renders/... 这条路径,而 renders 自身 只是一个指向 extracted 的目录级软链接。渲染文件实际都存放在 extracted/ 下, 所以 hypersimrenders 分组里除了那些文件,还显式收录了 renders 这个软链接本身; 少了它,新机上 renders/... 路径不存在,全部 images 链接都会断。 打包用 tar 不加 -h,保留链接本身;链接目标由对应分组整体收录。若展开链接, 同一张 HyperSim 渲染图会被数千个物体各拷一份,实测膨胀约 11 倍。 两条 run 使用相同数据。以 poserelabelAB 为例: 完整参数见各 run 目录下的 launch.sh。 训练用 --pointmap-source gt(默认)时,深度不是从物体目录读,而是由 scripts/hypersimso/rawdepth.py 按 cameras/.json 里的 camname / frameidx 实时拼出原始 HyperSim 路径: HYPERSIMRAWROOT 默认…
Independent publisher
Boingshaw
boingshaw
Models in Library0
Datasets in Library1
Models on Hugging Face—
Followers—
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Image to 3d.
Reported licenses: other. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.