Compressing Deep CNNs Using Basis Representation and Spectral Fine-Tuning

• Published in: 2021 IEEE International Conference on Image Processing (ICIP) pp. 3537 - 3541
• Main Authors: Tayyab, Muhammad, Khan, Fahad Ahmad, Mahalanobis, Abhijit
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.09.2021
• Subjects: Basis representation; Conferences; Convolution; Convolutional neural networks; Image classification; Image coding; network compression; Object detection; orthogonal filters; Spatial filters
• DOI: 10.1109/ICIP42928.2021.9506128

We propose an efficient and straightforward method for compressing deep convolutional neural networks (CNNs) that uses basis filters to represent the convolutional layers, and optimizes the performance of the compressed network directly in the basis space. Specifically, any spatial convolution layer of the CNN can be replaced by two successive convolution layers: the first is a set of three-dimensional orthonormal basis filters, followed by a layer of one-dimensional filters that represents the original spatial filters in the basis space. We jointly fine-tune both the basis and the filter representation to directly mitigate any performance loss due to the truncation. Generality of the proposed approach is demonstrated by applying it to several well known deep CNN architectures and data sets for image classification and object detection. We also present the execution time and power usage at different compression levels on the Xavier Jetson AGX processor.