Skip to content

Latest commit

 

History

History
executable file
·
59 lines (54 loc) · 3.5 KB

DATASET.md

File metadata and controls

executable file
·
59 lines (54 loc) · 3.5 KB

SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation

Open6DOR V2

Download the open6dor_v2.zip and extract it to ./datasets/open6dor_v2/. The overall directory structure should be:

│SoFar/datasets/open6dor_v2/
├── task_refine_pos/
│   ├── behind
│   └── ...
├── task_refine_rot/
│   └── rot_ins
└── task_refine_6dof/
    ├── behind
    │   ├── Place_the_apple_behind_the_bottle_on_the_table.__upright
    │   └── ...
    └── ...
Data file name Size
open6dor_v2.zip 1.89 GB
task_refine_pos 20.41 GB
task_refine_rot 12.84 GB
task_refine_6dof 21.99 GB

6-DoF SpatialBench

Download the images.zip and extract it to ./datasets/6dof_spatialbench/. The overall directory structure should be:

│SoFar/datasets/6dof_spatialbench/
├── spatial_data.json
└── images/
    ├── 0.png
    └── ...
Data file name Size
images.zip 531.7 MB
spatial_data.json 0.8 KB

OrienText300K

Download the OrienText300K.json or OrienText300K_strict.json and place it in ./orientation/data/SO/train.json. The strict version is a subset of the original dataset with more strict filtering, including smooth and texture surface. We use Cap3D as our point clouds data source. Cap3D_pcs data needs to be obtained in pt form from the Cap3D repository. The overall directory structure should be:

│orientation/data/SO/
├── train.json <---- OrienText300K.json
└── pcs <---- cap3d_pcs
    ├── 00000054c36d44a2a483bdbff31d8edf.pt
    └── ...
Data file name Size
OrienText300K.json 189 MB
OrienText300K_strict.json 140 MB
cap3d_pcs 173.8 GB