Convolutional Networks with Dense Connectivity

2020/01/08 by Gao Huang, Huang, Gao, Zhuang Liu +7 · 9 citations
Computer Science · Engineering · Mathematics · #Advanced Neural Network Applications #Algorithm #Artificial intelligence #Benchmark (surveying) #Computation #Computer Vision and Pattern Recognition (cs.CV) #Computer network #Computer science #Convolutional neural network #Domain Adaptation and Few-Shot Learning #Engineering #FOS: Computer and information sciences #Feature (linguistics) #Geology #Human Pose and Action Recognition #Layer (electronics) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Network architecture #Object (grammar) #Pattern recognition (psychology) #Reuse #cs.CV #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.2001.02394

published in arXiv (Cornell University) (Cornell University) · Journal(PAMI) version of DenseNet(CVPR'17)

arxiv created 2020/01/08 · openalex publication_date 2020/01/08 · arxiv updated 2020/01/09 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/08

Abstract

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion.Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, encourage feature reuse and substantially improve parameter efficiency. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less parameters and computation to achieve high performance.

Citations

• Semi-Supervised Learning with Ladder Networks
• Training Very Deep Networks
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• Gradient-based learning applied to document recognition
• ImageNet Large Scale Visual Recognition Challenge
• Striving for Simplicity: The All Convolutional Net
• Backpropagation Applied to Handwritten Zip Code Recognition
• ImageNet classification with deep convolutional neural networks
• Bridging the Gaps Between Residual Learning, Recurrent Neural Networks and Visual Cortex
• Training Deep Nets with Sublinear Memory Cost
• Wide Residual Networks
• AdaNet: Adaptive Structural Learning of Artificial Neural Networks
• Log-DenseNet: How to Sparsify a DenseNet
• Xception: Deep Learning with Depthwise Separable Convolutions
• Learning Multiple Layers of Features from Tiny Images

Cited by

• Towards Accurate and Efficient Waste Image Classification: A Hybrid Deep Learning and Machine Learning Approach
• SHREC'25 Track on Multiple Relief Patterns: Report and Analysis
• Characterizing the Reynolds number dependence of the chaotic attractor in two-dimensional turbulence with dimension-minimizing autoencoders
• ATR-Bench: A Federated Learning Benchmark for Adaptation, Trust, and Reasoning
• MWPRFN: Multilevel Wavelet Pyramid Recurrent Fusion Network for underwater image enhancement
• Applications of Deep Neural Networks with Keras

Related

• Densely Connected Convolutional Networks
• Connection Reduction of DenseNet for Image Recognition
• Exploring Feature Reuse in DenseNet Architectures
• Striving for Simplicity: The All Convolutional Net
• DenseNets Reloaded: Paradigm Shift Beyond ResNets and ViTs