N-EPIC-Kitchens: is the first event-based dataset for egocentric action recognition, presented at CVPR2022 in the paper "E^2(GO) MOTION: Motion Augmented Event Stream for Egocentric Action Recognition" Publication

Three largest kitchen of EPIC Kitchens-55 dataset (P01-P08-P22) are enanched with the event modality, following the adaptive sampling procedure proposed in Vid2E Then, using a framelike event encoding technique, called Voxel Grid, the sparse and asynchronous events are converted in a tensor representation enabling the learning with typical convolutional neural network architectures.

You can used the download script provided in the file download_data.sh or using rsync in the following way:

rsync -avhP rsync://vandaldata.polito.it/N-EPIC-Kitchens/  <path_to_save>

For only voxel_grid

rsync -avhP rsync://vandaldata.polito.it/N-EPIC-Kitchens/voxels_xy_3 <path_to_save>

For only raw event data

rsync -avhP rsync://vandaldata.polito.it/N-EPIC-Kitchens/events <path_to_save>

The directory structure:

├── voxels_xy_3/
|   
|   ├── P01_01
|   |   ├── event_0000000000.npy
|   |   ├── ...
|   ├── P01_02
|   ├── ...
| 
|   ├── P08_01
|   ├── ...
|   
|   ├── P22_01
|   ├── ...

If the entire download of voxels gets broken along the process and you only want to download certain videos to complete the dataset you can visit the Dropbox directory and select the files that you need.

Chiara Plizzari - email: chiara.plizzari@polito.it
Mirco Planamente - email: mirco.planamente@polito.it
Gabriele Goletto - email: gabriele.goletto@polito.it
Marco Cannici - email: marco.cannici@polimi.it

If this dataset was utilised, please cite:

@article{plizzari20212,
  title={E $\^{} 2$(GO) MOTION: Motion Augmented Event Stream for Egocentric Action Recognition},
  author={Plizzari, Chiara and Planamente, Mirco and Goletto, Gabriele and Cannici, Marco and Gusso, Emanuele and Matteucci, Matteo and Caputo, Barbara},
  journal={arXiv preprint arXiv:2112.03596},
  year={2021}
}