Skip to content

V1.0.0
Compare
Choose a tag to compare
@mwray mwray released this 01 Jul 10:11
· 36 commits to master since this release
v1.0.0
b33ca89

V1.0.0

First release of the annotations for the EPIC-KITCHENS-100 dataset.

This includes annotation files for the three main splits:

  • EPIC_100_train.csv for the training set.
  • EPIC_100_validation.csv for the validation set.
  • EPIC_100_test_timestamps.csv for the testing set.

Note that all narrations in any of the test sets do not have labelled verb/noun classes for the challenges.

As well as the verb/noun class files:

  • EPIC_100_verb_classes.csv for the verb classes and categories.
  • EPIC_100_noun_classes.csv for the noun classes and categories.

The annotations for the Unsupervised Domain Adaptation (UDA) and Action retrieval can be respectively found under UDA_annotations and retrieval_annotations.

See the readme.md for more information about the annotations.

We also provide pickle files for the annotations csvs. The version of the python pickles can be checked as follows:

>>> import pandas as pd
>>> train = pd.read_pickle('EPIC_100_train.pkl')
>>> train._metadata['version_number']
'v1.0.0'

There is also a commit number (accessible via train._metadata['commit_hash']) should you need to check specifically which commit these pickles were generated from.

Assets 2
Loading