Event-based Dataset of Assembly Tasks (EDAT24)
Description
Version 2:
Re-upload to replace empty .npy files.
IMPORTANT NOTE: pick/bridge_roof_2, pick/bridge_roof_3 and pick/bridge_roof_4 files have background movement (a walking person) and, as such, should be used carefully when deploying the dataset.
------------------------------------------------------------------------------------------------------------------------------------------------
Description:
The Event-based Dataset of Assembly Tasks (EDAT24), features four representative assembly primitive actions, idle, pick, place, and screw, collected by using the objects from the open-source benchmark CT-Benchmark. The dataset features a total of 400 recordings, with 100 samples for each primitive action.
Available Data:
The EDAT24 dataset contains a total of 400 sample videos recorded with the DAVIS240C where the event camera captured events and APS frames simultaneously with a resolution of 240 pixels × 180 pixels. DAVIS APS data are captured at 20 fps (frames per second). Event data is available in the original recorded format, .aedat (AEDAT 2.0), and is accompanied with a .csv file containing the timestamps with the start and end time of the recording, and with a .npy file which contains the information from the events (coordinates x and y, timestamp and polarity).
The videos are labelled using the format primitive/task_nSample.aedat, where:
- primitive corresponds to the various primitive tasks (idle, pick, place, and screw)
- task corresponds to the name of the part of the CT-Benchmark the user is interacting with.
- nSample corresponds to the sample number for each task.
The Python code used to record, timestamp and convert into .npy is available at here.
Files
idle.zip
Additional details
Related works
- Is continued by
- Dataset: 10.5281/zenodo.15005301 (DOI)
- Is described by
- Journal article: 10.1016/j.dib.2024.110340 (DOI)
- Is supplemented by
- Software: https://github.com/Robotics-and-AI/DAVIS-data-capture-system (URL)