The 2nd WEAR Dataset Challenge is a Human Activity Recognition prediction challenge based on the inertial data of the WEAR dataset. The challenge will be again part of the HASCA Workshop at UbiComp/ ISWC 2025.
This year's challenge will be all about robustness and generalization! We curated a new test dataset with four new participants.
But there is a twist! This year we only provide you with random, sensor-specific 1-second sliding windows of each of the four test participants. Oh yeah, and we applied some random augmentations to the data that can occur due to different wearing conditions. This means that you will have to rely on your model's ability to generalize to new participants and deal with the noise and uncertainty that comes with real-world data.
More details can be found on the Kaggle Challenge website.
We are happy to announce that the WEAR Challenge is now also available on Kaggle! You can find the challenge here. Unlike the first iteration of the challenge, submission of prediction results will only be done via Kaggle. By switching to Kaggle we hope to make the challenge more accessible to a wider audience and provide a more streamlined submission process. By using Kaggle's public and private leaderboard we hope to provide a more transparent and fair ranking process. Challenge participants will further be able to discuss the challenge and share their solutions on the Kaggle platform.
We will announce winners of the challenge at the HASCA workshop. In order to be eligible for prizes and be part of the final ranking announced at the conference, participants need to tell us when submitting their technical report, which submission on Kaggle is their final submission. Prizes will be awarded to the three teams with the highest private leaderboard score at the time of the paper submission deadline.
The WEAR dataset is an outdoor sports dataset for inertial- and video-based human activity recognition (HAR). The dataset comprises data from 22 participants performing a total of 18 different workout activities with untrimmed inertial (acceleration) and egocentric video data recorded at 11 different outside locations. WEAR provides a challenging prediction scenario marked by purposely introduced activity variations as well as an overall small information overlap across modalities. This challenge focuses on the inertial data only.
Please visit the Kaggle challenge page to download the training and test data. The training data is provided as a directory containing the CSV files of the 22 participants. The test data is provided as a single CSV file with each row representing one of the sliding windows of the 4 new participants, which are to be predicted. For more details please refer to the Data section on Kaggle.
Evaluation of submissions will be based on the macro F1-score. The macro F1-score is the average of the F1-scores of each activity class.
Submission of the predictions will happen solely via Kaggle. All details on the submission format and final submissions can be found on Kaggle.
The final ranking of the WEAR Dataset challenge will be announced at the HASCA workhop. Ranking of the teams, which will be eligible for prizes, will be determined based on the private leaderboard score as of the technical report submission deadline.
When submitting their technical report, participants will need to:
WEAR is offered under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. You are free to use, copy, and redistribute the material for non-commercial purposes provided you give appropriate credit, provide a link to the license, and indicate if changes were made. If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original. You may not use the material for commercial purposes.