datasets.md

June 13, 2024 · View on GitHub

:floppy_disk: Datasets

Follow the instructions below to setup the datasets and assets.
 

DexYCB

Download DexYCB dataset from the official site, unzip and link the dataset to ./data/DexYCB.
The ./data/DexYCB directory should have the following structure:

├── DexYCB
│   ├── 20200709-subject-01
│   ├── ...
│   ├── 20201022-subject-10
│   ├── bop
│   ├── calibration
│   └── models

HO3D v3

Download HO3D_v3 from the official site, unzip and link the dataset to ./data/HO3D_v3.
The ./data/HO3D_v3 should have the following structure:

├── HO3D_v3
│   ├── calibration
│   ├── evaluation
│   ├── evaluation.txt
│   ├── manual_annotations
│   ├── train
│   └── train.txt

:bell: Prepare HO3D-MV (HO3D_v3 in multi-view)

Sequence: 'GPMF1' and 'SB1' in HO3D v3's official testing set are included by HO3D-MV dataset. We provide ground-truth MANO annotation for frames in 'GPMF1' and 'SB1' sequences.

Download the :link: HO3D_v3_manual_test_gt.zip, unzip, and link the folder to ./data/HO3D_v3_manual_test_gt.

├── HO3D_v3
+── HO3D_v3_manual_test_gt

OakInk

Download OakInk dataset from the official site, unzip and link the dataset in ./data/OakInk.
The ./data/OakInk directory should have the following structure:

├── OakInk
│   ├── OakBase
│   ├── image
│   │   ├── anno
│   │   ├── obj
│   │   └── stream_release_v2
│   └── shape
│       ├── metaV2
│       ├── OakInkObjectsV2
│       ├── oakink_shape_v2
│       └── OakInkVirtualObjectsV2

:bell: Prepare OakInk-MV (OakInk in multi-view)

Step 1. extend the OakInk data splits for multi-view setting:

$ sh prepare/extend_oakink_mv_splits.sh

Step 2. pack annotations into a single archive for each sample (for quick loading):

$ sh prepare/pack_oakink_mv_anno.sh

After the above steps, the ./data/OakInk directory should have the following structure:

├── OakInk
│   ├── OakBase
│   ├── image
│   │   ├── anno
+   │   ├── anno_mv
+   │   ├── anno_packed_mv
│   │   ├── obj
│   │   └── stream_release_v2
│   └── shape

:heavy_check_mark: After all three datasets have been successfully processed, you will have ./data directory as:

├── data
│   ├── DexYCB
│   ├── HO3D_v3
│   ├── HO3D_v3_manual_test_gt
│   └── OakInk

 

:luggage: Assets

Get the MANO hand model mano_v1_2.zip from the MANO website.

  1. click Download on the top menu, this requires register & login.
  2. on the Download page, navigate to Models & Code section, and click Models & Code, the mano_v1_2.zip will be downloaded automatically.
  3. Unzip mano_v1_2.zip and copy it into an assets folder.
$ mkdir assets
$ cp -r {path/to}/mano_v1_2 ./assets/