README.md

Data preparation and directory structure

We now provide the instructions to populate the datasets under data/. Make sure that all data is arranged using the provided directory structure.

data
├── amazon-berkeley
│   └── 3dmodels
│       └── original
├── datasets
│   └── cos_eor_v11_pruned
│       ├── train
│       ├── val
│       └── test
├── default.physics_config.json  
├── google_object_dataset
│       ├── 11pro_SL_TRX_FG
│       └── ...
├── objects
│   ├── configs_convex
│   ├── ycb
│   ├── 002_heavy_master_chef_can.object_config.json
│   └── ...
├── replica-cad
│   ├── configs
│   ├── objects
│   └── ...
├── scene_datasets
│   └── igibson
│       ├── assets
│       ├── assets_assemble
│       └── scenes
└── urdf_obj_cache

Amazon Berkeley Objects

• Download and extract the Amazon Berkeley 3D models dataset (abo-3dmodels.tar) inside data/amazon-berkeley.
• Run python cos_eor/scripts/ab_objects_viz.py to generate config files for the downloaded objects

Google Scanned Objects

• Run the script python cos_eor/scripts/gso_downloader.py to download the Google Scanned Objects and generate config files.

YCB objects

• Run the script ./cos_eor/scripts/ycb/full_setup.sh. This script will download YCB objects under data/objects, generate physics config files and downscale the assets.

Replica-CAD dataset

• Download ReplicaCAD Interactive objects dataset under data/replica-cad and extract it to data/replica-cad.

Downscaling the assets

• Use the script python cos_eor/scripts/downscale_assets.py to downscale the downloaded assets.