BABEL

BABEL Action Recognition Challenge

Evaluation Server: https://babel-evaluation.is.tuebingen.mpg.de/

Use the code in action_recognition/challenge/create_submission.py, in our Github repo to create a submission file.

Task

Given a motion-capture span of 5 sec. or lesser, predict the likelihood (score) of actions in it. See the BABEL paper for details regarding the task.

See the action_recognition/ folder in our Github repo for reference code for a baseline model.

Challenge settings

There are 4 different “Challenge settings” in the action recognition task, with two variables:

Number of classes

BABEL-60 subset containing spans of mocap sequences belonging to 60 classes.
BABEL-120 subset containing spans of mocap sequences belonging to 120 classes.

Type of labels

Dense labels only — the submission uses only the following two files: train.json and val.json for training and validation.
Dense + Extra labels — the submission uses train.json, val.json, and in addition, extra_train.json and extra_val.json during training and validation.

For details regarding the Dense labels and Extra labels, please visit our Data page.

Predictions

Please refer to the action_recognition/challenge folder in the BABEL GitHub repository for details regarding generating a submission file on the test set.

Submit your predictions to our Action Recognition Evaluation Server (Coming Soon).

Acknowledgement

Thanks to the Software Workshop at Max-Planck Institute for Intelligent Systems for developing the BABEL Action Recognition evalation server.