Multi-task, multi-domain learning: Application to semantic segmentation and pose regression

NEUROCOMPUTING

卷:251

Article

Fourure, Damien^1,2  Emonet, Remi¹  Fromont, Elisa¹  Muselet, Damien¹  Neverova, Natalia²  Tremeau, Alain¹  Wolf, Christian² 

[1] Univ Lyon, UJM, CNRS, Lab Hubert Curien UMR5516, F-42000 Lyon, France

[2] Univ Lyon, CNRS, INSA Lyon, LIRIS,UMR5205, F-69621 Lyon, France

关键词: Deep learning;    Convolutional neural networks;    Semantic segmentation;    Domain adaptation;    Multi-task learning;   

DOI  :  10.1016/j.neucom.2017.04.014

来源: Elsevier

【 摘 要 】

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 labelsets. 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 segmentation and hand pose estimation) show the relevance of our approach in different contexts. (C) 2017 Elsevier B.V. All rights reserved.

【 授权许可】

Free