README.md

metadata

license: mit
configs:
  - config_name: default
    data_files:
      - split: train
        path: episodes.csv

Human Pick-and-Place Dataset

About

This dataset contains 100 episodes of humans performing the pick-and-place task. The data captured consists of multiple modalities. These include rgb-color footage, depth footage and gyro-accelerometer data captured using a head mounted camera-IMU unit, as well as full-body motion capture data recorded by an inertial-based MoCap suit.

Folder Structure

├── episodes.csv
├── [color]
│   ├── [MP4 file]
│   ├── [MP4 file]
│   ├── [MP4 file]
│   └── ...
├── [depth]
│   ├── [JXL file]
│   └── ...
└── ...

The above diagram shows the directory layout for this dataset. episodes.csv contains high level information on the dataset while the data files are organized into different folders based on modalities.

episodes.csv

Contains metadata and statistics for this dataset. episodes.csv has the following columns (JSONs are semicolon separated):

• episode_id: string unique identifier for episode
• period: int total period of episode in seconds (rounded to an integer)
• num_modalities: int number of modalities of data associated with the episode
• metadata: JSON relevant metadata regarding the episode
• modalities: JSON information about the different modalities of data collected View the file in dataviewer for more details.

Data files

Data files are organized into different folder based on modalities. Each data file is named after an episode_id, which refers to the data file's corresponding episode.