Deep saliency: What is learnt by a deep network about saliency?
International Machine Learning Society
Copyright 2017 by the author(s).
Reason for embargo
Under embargo until after conference
Deep convolutional neural networks have achieved impressive performance on a broad range of problems, beating prior art on estab- lished benchmarks, but it often remains unclear what are the representations learnt by those systems and how they achieve such performance. This article examines the specific problem of saliency detection, where benchmarks are currently dominated by CNN-based approaches, and investigates the properties of the learnt rep- resentation by visualizing the artificial neurons’ receptive fields. We demonstrate that fine tuning a pre-trained network on the saliency detection task lead to a profound transformation of the network’s deeper layers. Moreover we argue that this transformation leads to the emergence of receptive fields conceptually similar to the centre-surround filters hypothesized by early research on visual saliency.
This work was supported by the EPSRC project DEVA EP/N035399/1.
2nd Workshop on Visualization for Deep Learning
This is the author accepted manuscript. The final version is available from the International Machine Learning Society via the URL in this record.
34th International Conference on Machine Learning, 6-11 August 2017, Sidney, Australia