The repository contans source code for labeljitter paper from ISWC 2019
Script for Random Forest - SAT: main_rf.py The script shows how to use SAT for the classifier that only allows discrete label.
Script for ConvLSTM - SAT: main_dl.py The (keras) script shows how to use SAT for the classifier that can train continuous labels end-to-end.
The necessary datasets and trained models used for the paper can be downloaded from following project webpage: https://sites.google.com/view/labeljitteriswc19
Direct link to the paper: https://dl.acm.org/citation.cfm?doid=3341163.3347744
Direct link to the datasets: https://www.dropbox.com/s/zpohgbm2622l6pm/dataset.zip?dl=0
Direct link to the trained model: https://www.dropbox.com/s/9w3pg4frslxt8vq/models.zip?dl=0
When you use any resources (code, traned models) related to label jitter project, please cite following paper.
Kwon, H., Abowd, G. D., & Plötz, T. (2019, September). Handling Annotation Uncertainty in Human Activity Recognition In Proceedings of the 2019 ACM International Symposium on Wearable Computers. ACM.