IndustReal Dataset of Egocentric Videos for Procedure Understanding

Datacite citation style:
Schoonbeek, Tim J.; Houben, Tim; Onvlee, Hans; de With, Peter H.N.; van der Sommen, Fons (2023): IndustReal Dataset of Egocentric Videos for Procedure Understanding. Version 1. 4TU.ResearchData. dataset. https://doi.org/10.4121/b008dd74-020d-4ea4-a8ba-7bb60769d224.v1
Other citation styles (APA, Harvard, MLA, Vancouver, Chicago, IEEE) available at Datacite
Dataset
choose version: version 2 - 2024-08-23 (latest)
version 1 - 2023-10-31

The multi-modal IndustReal dataset, accompanying our publication "IndustReal: A Dataset for Procedure Step Recognition Handling Execution Errors in Egocentric Videos in an Industrial-Like Setting". Check out our GitHub for additional details and read-me files.


Unlike currently available datasets, IndustReal contains procedural errors (such as omissions) as well as execution errors. A significant part of these errors are exclusively present in the validation and test sets, making IndustReal suitable to evaluate robustness of algorithms to new, unseen mistakes. Additionally, to encourage reproducibility and allow for scalable approaches trained on synthetic data, the 3D models of all parts are publicly available.

history
  • 2023-10-31 first online, published, posted
publisher
4TU.ResearchData
format
video frames (jpg), hand, pose and gaze tracking (csv), ground-truth labels (csv, json), object geometries (fbx). Model weights (.PYTH and .pt) and predictions (.csv) also provided. Everything is contained in zip or tar files.
organizations
Eindhoven University of Technology, Department of Electrical Engineering

DATA

files (18)