Ronaldo-GOAT/ho3d_transfer

HO-3D v3 (merged transfer)

This repository bundles the HO-3D v3 dataset together with its rendered hand/object segmentations, kept as the original .zip archives plus the upstream README.txt.

Contents

File	Size	Description
HO3D_v3.zip	~32 GB	Main dataset: train/, evaluation/, calibration/, manual_annotations/, train.txt, evaluation.txt
HO3D_v3_segmentations_rendered.zip	~91 MB	Rendered hand/object segmentations: train/<SEQ>/seg/*.png (1/4 RGB resolution)
README.txt	—	Original upstream documentation (description, annotation format, citation)

Merging into a single folder

The segmentation archive shares the same train/<SEQ>/ layout as the main archive, so the two extract and merge cleanly (the seg/ subfolders are new — no file conflicts):

unzip HO3D_v3.zip -d HO3D_v3
unzip HO3D_v3_segmentations_rendered.zip -d HO3D_v3   # adds train/<SEQ>/seg/

License / terms

Academic, non-commercial research use only. The upstream terms state the dataset may not be redistributed, modified, or used commercially without the authors' express permission. See README.txt for the full terms.

Citation

@ARTICLE{hampali2019honnotate,
  title   = {HOnnotate: A method for 3D Annotation of Hand and Object Poses},
  author  = {Shreyas Hampali and Mahdi Rad and Markus Oberweger and Vincent Lepetit},
  booktitle = {Proc. Computer Vision and Pattern Recognition (CVPR), IEEE},
  year    = {2020}
}