OpenPack Dataset Toolkit (optk)

"OpenPack Dataset" is a new large-scale multi modal dataset of manual packing process. OpenPack is an open access logistics-dataset for human activity recognition, which contains human movement and package information from 16 distinct subjects. This repository provide utilities to explore our exciting dataset.

Dataset Release Note

• OpenPack Dataset (v1.0.0)

For preliminary analysis, please start from preprocessed-IMU-with-operation-labels.zip in zenodo. This preprocessed dataset include IMU data (acc, gyro, quaternion) assosiated with operatopn labels because you don't need to combine data and label.

Docs

Dataset

• Subjects & Recording Scenarios
• Activity Class Definition
• Modality
• Data Split (Train/Val/Test/Submission)

Task & Activity Recognition Challenge

• Work Operation Recognition
  • OpenPack Challenge 2022

Sample Data

Sample

Install

We provide some utility functions as python package. You can install via pip with the following command. Note that the supported dataset version is >=1.0.0.

# Pip
pip install openpack-toolkit

# Poetry
poetry add  openpack-toolkit

Examples

Tutorial

• Download and Visualization (Notebook)

Work Activity Recognition (PyTorch)

PyTorch code samples for work operation prediction task is available. See openpack-torch for more dietail.

Timestamp

Each data point is associated with a millisecond-precision unix timestamp. The following is a snippet that converts a timestamp (an int value) into a datatime.datetime() object with timezone.

import datetime


def timestamp_to_datetime(ts: int) -> datetime.datetime:
  """Convert unix timestamp (milli-second precision) into datatime object. """
  timezone_jst = datetime.timezone(datetime.timedelta(hours=9))
  dt = datetime.datetime.fromtimestamp(ts / 1000).replace(tzinfo=timezone_jst)
  return dt
  
def datetime_to_timestamp(dt: datetime.datetime) -> int:
  """Convert a datetime object into a milli-second precision timestamp."""
  return int(dt.timestamp() * 1000)


ts = 1634885786000

dt_out =  timestamp_to_datetime(ts)
ts_out = datetime_to_timestamp(dt_out)
print(f"datetime: {dt_out}")  # datetime: 2021-10-22 15:56:26+09:00
print(f"timestamp: {ts_out}")  # timestamp: 1634885786000
assert ts_out == ts

Download Dataset

From Zenodo (w/o Depth Images)

bash tools/download/dl_from_zenodo.sh <path to a dataset root directory>

# Example:
bash tools/download/dl_from_zenodo.sh ./data/datasets

Links

• Homepage (External Site)
  • OpenPack Challenge 2022 (External Site)
• zenodo
• API Docs
• PyPI
• openpack-torch

License

openpack-toolkit has a MIT license, as found in the LICENSE file.

NOTE: OpenPack Dataset itself is available under Creative Commons Attribution Non Commercial Share Alike 4.0 International license. However, OpenPack Dataset (+RGB) License is applied to "OpenPack Dataset (+RGB)" which includs RGB data.