DSConv: efficient convolution operator

Conference item

Quantization is a popular way of increasing the speed and lowering the memory usage of Convolution Neural Networks (CNNs). When labelled training data is available, network weights and activations have successfully been quantized down to 1-bit. The same cannot be said about the scenario when labelled training data is not available, e.g. when quantizing a pre-trained model, where current approaches show, at best, no loss of accuracy at 8-bit quantizations. We introduce DSConv, a flexible quantized convolution operator that replaces single-precision operations with their far less expensive integer counterparts, while maintaining the probability distributions over both the kernel weights and the outputs. We test our model as a plug-and-play replacement for standard convolution on most popular neural network architectures, ResNet, DenseNet, GoogLeNet, AlexNet and VGG-Net and demonstrate state-of-the-art results, with less than 1% loss of accuracy, without retraining, using only 4-bit quantization. We also show how a distillation-based adaptation stage with unlabelled data can improve results even further.

Authors: Gennari do Nascimento, M; Prisacariu, V; Fawcett, R

• Alternative title: Conference paper
• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Institute of Electrical and Electronics Engineers
• Host title: 2019 IEEE/CVF International Conference on Computer Vision (ICCV)
• Pages: 5147-5156
• Publication date: 2019-02-27
• Acceptance date: 2020-07-27
• Event title: 2019 IEEE/CVF International Conference on Computer Vision (ICCV)
• Event location: Seoul, South Korea
• Event start date: 2019-10-27
• Event end date: 2019-11-02
• DOI: 10.1109/ICCV.2019.00525
• EISSN: 2380-7504
• ISSN: 1550-5499
• EISBN: 9781728148038
• ISBN: 9781728148045
• Language: English
• Keywords: data models; neural networks; tensile stress; training data; quantization (signal); kernel; FFR; convolution