Multi-task, Multi-domain Learning: application to semantic segmentation and pose regression

Abstract : We present an approach that leverages multiple datasets annotated for different tasks (e.g., classification with different labelsets) to improve the predictive accuracy on each individual dataset. Domain adaptation techniques can correct dataset bias but they are not applicable when the tasks differ, and they need to be complemented to handle multi-task settings. We propose a new selective loss function that can be integrated into deep neural networks to exploit training data coming from multiple datasets annotated for related but possibly different label sets. We show that the gradient-reversal approach for domain adaptation can be used in this setup to additionally handle domain shifts. We also propose an auto-context approach that further captures existing correlations across tasks. Thorough experiments on two types of applications (semantic segmenta-tion and hand pose estimation) show the relevance of our approach in different contexts.
Complete list of metadatas

Cited literature [40 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01507132
Contributor : Damien Fourure <>
Submitted on : Tuesday, July 4, 2017 - 12:47:42 PM
Last modification on : Tuesday, February 26, 2019 - 4:35:37 PM
Long-term archiving on : Thursday, December 14, 2017 - 11:34:34 PM

File

Multi-task_Multi-domain_Learni...
Files produced by the author(s)

Identifiers

Citation

Damien Fourure, Rémi Emonet, Elisa Fromont, Damien Muselet, Natalia Neverova, et al.. Multi-task, Multi-domain Learning: application to semantic segmentation and pose regression. Neurocomputing, Elsevier, 2017, 251, pp.68-80. ⟨10.1016/j.neucom.2017.04.014⟩. ⟨hal-01507132v2⟩

Share

Metrics

Record views

304

Files downloads

796