Robot Locomotion Group Manipulation Dataset

This repo contains instructions on how to download the datasets used in the following papers.

• DenseObjectNets
• kPAM

DenseObjectNets

To download the Dense Object Nets data you can use

python3 data_downloader.py --config_file "config/all_episodes.yaml"

Note this downloads a large number of episodes. Create your own config if you only want to download a subset.

kPAM Data

Annotations

Can be downloaded using this link.

Episode Visual Data

The visual data for kPAM consists of episodes. See the "Downloading Specific Episodes" section for more details. The config files config/kpam_mugs.yaml and config/kpam_shoes.yaml contain the episodes used for shoes and mugs repsectively in the kPAM paper.

Mugs

To download the mugs data use

python3 data_downloader.py --config_file "config/kpam_mugs.yaml"

Shoes

To download the shoe data use

python3 data_downloader.py --config_file "config/kpam_shoes.yaml"

Keypoints into the Future

The episodes used to train the dense-descriptors model can be downloaded with

python3 data_downloader.py --config_file "config/kitf_hardware_perception.yaml"

The episodes that are used for dynamics learning can be downloaded with

python3 data_downloader.py --config_file "config/kitf_hardware_dynamics.yaml"

Downloading Specific Episodes

An episode (also known as a scene) consists of many images for which relative camera poses are known. See here for more information. A list of episodes can be found in config/all_episode.yaml. Downloading a list of episodes specified in a yaml file can be accomplished by running

python3 data_downloader.py --config_file <full_path_to_config> --dest <full_path_to_dest_dir>  

This will download the episodes the specified destination directory. For example to download the entire dataset use

python3 data_downloader.py --config_file "config/all_episodes.yaml"