Transferring and compressing convolutional neural networks for face representations
(2016) 13th International Conference on Image Analysis and Recognition, ICIAR 2016 In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 9730. p.20-29- Abstract
In this work we have investigated face verification based on deep representations from Convolutional Neural Networks (CNNs) to find an accurate and compact face descriptor trained only on a restricted amount of face image data. Transfer learning by fine-tuning CNNs pre-trained on large-scale object recognition has been shown to be a suitable approach to counter a limited amount of target domain data. Using model compression we reduced the model complexity without significant loss in accuracy and made the feature extraction more feasible for real-time use and deployment on embedded systems and mobile devices. The compression resulted in a 9-fold reduction in number of parameters and a 5-fold speed-up in the average feature extraction... (More)
In this work we have investigated face verification based on deep representations from Convolutional Neural Networks (CNNs) to find an accurate and compact face descriptor trained only on a restricted amount of face image data. Transfer learning by fine-tuning CNNs pre-trained on large-scale object recognition has been shown to be a suitable approach to counter a limited amount of target domain data. Using model compression we reduced the model complexity without significant loss in accuracy and made the feature extraction more feasible for real-time use and deployment on embedded systems and mobile devices. The compression resulted in a 9-fold reduction in number of parameters and a 5-fold speed-up in the average feature extraction time running on a desktop CPU. With continued training of the compressed model using a Siamese Network setup, it outperformed the larger model.
(Less)
- author
- Grundström, Jakob ; Chen, Jiandan LU ; Ljungqvist, Martin Georg and Åström, Kalle LU
- organization
- publishing date
- 2016
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- series title
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- volume
- 9730
- pages
- 10 pages
- publisher
- Springer
- conference name
- 13th International Conference on Image Analysis and Recognition, ICIAR 2016
- conference location
- Povoa de Varzim, Portugal
- conference dates
- 2016-07-13 - 2016-07-16
- external identifiers
-
- wos:000386604000003
- scopus:84978872481
- ISSN
- 1611-3349
- 0302-9743
- ISBN
- 978-3-319-41501-7
- 978-3-319-41500-0
- DOI
- 10.1007/978-3-319-41501-7_3
- language
- English
- LU publication?
- yes
- id
- 04a61dac-b616-4585-8046-f35e569f7933
- date added to LUP
- 2016-08-29 08:37:06
- date last changed
- 2024-06-28 14:08:56
@inproceedings{04a61dac-b616-4585-8046-f35e569f7933, abstract = {{<p>In this work we have investigated face verification based on deep representations from Convolutional Neural Networks (CNNs) to find an accurate and compact face descriptor trained only on a restricted amount of face image data. Transfer learning by fine-tuning CNNs pre-trained on large-scale object recognition has been shown to be a suitable approach to counter a limited amount of target domain data. Using model compression we reduced the model complexity without significant loss in accuracy and made the feature extraction more feasible for real-time use and deployment on embedded systems and mobile devices. The compression resulted in a 9-fold reduction in number of parameters and a 5-fold speed-up in the average feature extraction time running on a desktop CPU. With continued training of the compressed model using a Siamese Network setup, it outperformed the larger model.</p>}}, author = {{Grundström, Jakob and Chen, Jiandan and Ljungqvist, Martin Georg and Åström, Kalle}}, booktitle = {{Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}}, isbn = {{978-3-319-41501-7}}, issn = {{1611-3349}}, language = {{eng}}, pages = {{20--29}}, publisher = {{Springer}}, series = {{Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}}, title = {{Transferring and compressing convolutional neural networks for face representations}}, url = {{http://dx.doi.org/10.1007/978-3-319-41501-7_3}}, doi = {{10.1007/978-3-319-41501-7_3}}, volume = {{9730}}, year = {{2016}}, }