cicyt UNIZAR
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: DeepN-JPEG: A Deep Neural Network Favorable JPEG-based Image Compression Framework

Abstract: As one of most fascinating machine learning techniques, deep neural network (DNN) has demonstrated excellent performance in various intelligent tasks such as image classification. DNN achieves such performance, to a large extent, by performing expensive training over huge volumes of training data. To reduce the data storage and transfer overhead in smart resource-limited Internet-of-Thing (IoT) systems, effective data compression is a "must-have" feature before transferring real-time produced dataset for training or classification. While there have been many well-known image compression approaches (such as JPEG), we for the first time find that a human-visual based image compression approach such as JPEG compression is not an optimized solution for DNN systems, especially with high compression ratios. To this end, we develop an image compression framework tailored for DNN applications, named "DeepN-JPEG", to embrace the nature of deep cascaded information process mechanism of DNN architecture. Extensive experiments, based on "ImageNet" dataset with various state-of-the-art DNNs, show that "DeepN-JPEG" can achieve ~3.5x higher compression rate over the popular JPEG solution while maintaining the same accuracy level for image recognition, demonstrating its great potential of storage and power efficiency in DNN-based smart IoT system design.
Comments: 55th Design Automation Conference (DAC2018)
Subjects: Computer Vision and Pattern Recognition (cs.CV); Graphics (cs.GR); Performance (cs.PF)
Cite as: arXiv:1803.05788 [cs.CV]
  (or arXiv:1803.05788v1 [cs.CV] for this version)

Submission history

From: Tao Liu [view email]
[v1] Wed, 14 Mar 2018 02:18:55 GMT (2767kb,D)