Structured Binary Neural Networks for Accurate Image Classification and Semantic Segmentation

2018/11/22 by Bohan Zhuang, Zhuang, Bohan, Chunhua Shen +7 · 2 citations
Computer Science · Mathematics · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Algorithm #Artificial intelligence #Artificial neural network #Binary number #Computation #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Contextual image classification #Convolutional neural network #Domain Adaptation and Few-Shot Learning #Embedding #FOS: Computer and information sciences #Image (mathematics) #Mathematics #Pattern recognition (psychology) #Quantization (signal processing) #Segmentation #Theoretical computer science #cs.CV

paper · pdf · doi:10.48550/arxiv.1811.10413

arXiv admin note: text overlap with arXiv:1808.02631

openalex publication_date 2018/11/22 · arxiv created 2018/11/27 · arxiv updated 2018/11/29 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

In this paper, we propose to train convolutional neural networks (CNNs) with both binarized weights and activations, leading to quantized models specifically for mobile devices with limited power capacity and computation resources. Previous works on quantizing CNNs seek to approximate the floating-point information using a set of discrete values, which we call value approximation, but typically assume the same architecture as the full-precision networks. In this paper, however, we take a novel 'structure approximation' view for quantization---it is very likely that a different architecture may be better for best performance. In particular, we propose a `network decomposition' strategy, named Group-Net, in which we divide the network into groups. In this way, each full-precision group can be effectively reconstructed by aggregating a set of homogeneous binary branches. In addition, we learn effective connections among groups to improve the representational capability. Moreover, the proposed Group-Net shows strong generalization to other tasks. For instance, we extend Group-Net for highly accurate semantic segmentation by embedding rich context into the binary structure. Experiments on both classification and semantic segmentation tasks demonstrate the superior performance of the proposed methods over various popular architectures. In particular, we outperform the previous best binary neural networks in terms of accuracy and major computation savings.

Citations

• Training Compact Neural Networks with Binary Weights and Low Precision Activations
• Adam: A Method for Stochastic Optimization
• BinaryConnect: Training Deep Neural Networks with binary weights during propagations
• The Pascal Visual Object Classes (VOC) Challenge
• ImageNet Large Scale Visual Recognition Challenge
• MXNet: A Flexible and Efficient Machine Learning Library for Heterogeneous Distributed Systems
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• Binarized Neural Networks
• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations
• Neural Architecture Search with Reinforcement Learning
• Trained Ternary Quantization
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• Network Sketching: Exploiting Binary Structure in Deep CNNs
• Rethinking Atrous Convolution for Semantic Image Segmentation
• Channel Pruning for Accelerating Very Deep Neural Networks
• Performance Guaranteed Network Acceleration via High-Order Residual Quantization
• Towards Effective Low-bitwidth Convolutional Neural Networks
• Progressive Neural Architecture Search
• Loss-aware Weight Quantization of Deep Networks
• Heterogeneous Bitwidth Binarization in Convolutional Neural Networks
• Model compression via distillation and quantization
• Rethinking the Inception Architecture for Computer Vision
• Deep Residual Learning for Image Recognition
• Going Deeper with Convolutions
• LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks
• Bi-Real Net: Enhancing the Performance of 1-bit CNNs With Improved Representational Capability and Advanced Training Algorithm
• Deep Learning with Low Precision by Half-wave Gaussian Quantization
• Xception: Deep Learning with Depthwise Separable Convolutions
• Fully Convolutional Networks for Semantic Segmentation
• Aggregated Residual Transformations for Deep Neural Networks

Cited by

• Neural Network Quantization for Microcontrollers: A Comprehensive Survey of Methods, Platforms, and Applications
• StaticSegFormer: An Efficient High-Performance Semantic Segmentation Based on Static Structured Pruning

Related

• Structured Binary Neural Networks for Image Recognition
• Training Compact Neural Networks with Binary Weights and Low Precision Activations
• Efficient Convolutional Neural Network with Binary Quantization Layer
• CNN-aware Binary Map for General Semantic Segmentation
• Training and Inference for Integer-Based Semantic Segmentation Network