MeliusNet: Can Binary Neural Networks Achieve MobileNet-level Accuracy?

Abstract

Binary Neural Networks (BNNs) are neural networks which use binary weights and activations instead of the typical 32-bit floating point values. They have reduced model sizes and allow for efficient inference on mobile or embedded devices with limited power and computational resources. However, the binarization of weights and activations leads to feature maps of lower quality and lower capacity and thus a drop in accuracy compared to traditional networks. Previous work has increased the number of channels or used multiple binary bases to alleviate these problems. In this paper, we instead present an architectural approach: MeliusNet. It consists of alternating a DenseBlock, which increases the feature capacity, and our proposed ImprovementBlock, which increases the feature quality. Experiments on the ImageNet dataset demonstrate the superior performance of our MeliusNet over a variety of popular binary architectures with regards to both computation savings and accuracy. Furthermore, with our method we trained BNN models, which for the first time can match the accuracy of the popular compact network MobileNet-v1 in terms of model size, number of operations and accuracy. Our code is published online at https://github.com/hpi-xnor/BMXNet-v2

References in corpus (14)

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
• SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
• Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation
• DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
• Binarized Neural Networks
• Trained Ternary Quantization
• GluonCV and GluonNLP: Deep Learning in Computer Vision and Natural Language Processing
• WRPN: Wide Reduced-Precision Networks
• XNOR-Net++: Improved Binary Neural Networks
• Moonshine: Distilling with Cheap Convolutions
• Training Binary Neural Networks with Real-to-Binary Convolutions
• Training Competitive Binary Neural Networks from Scratch
• Binary Ensemble Neural Network: More Bits per Network or More Networks per Bit?

Cited by in corpus (15)

• Rotated Binary Neural Network
• SLSNet: Skin lesion segmentation using a lightweight generative adversarial network
• ReActNet: Towards Precise Binary Neural Network with Generalized Activation Functions
• Training Binary Neural Networks using the Bayesian Learning Rule
• BoolNet: Minimizing The Energy Consumption of Binary Neural Networks
• Sharpness-aware Quantization for Deep Neural Networks
• Accelerating Binarized Neural Networks via Bit-Tensor-Cores in Turing GPUs
• Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation
• Differentiable Joint Pruning and Quantization for Hardware Efficiency
• S2-BNN: Bridging the Gap Between Self-Supervised Real and 1-bit Neural Networks via Guided Distribution Calibration
• SiMaN: Sign-to-Magnitude Network Binarization
• AskewSGD : An Annealed interval-constrained Optimisation method to train Quantized Neural Networks
• FTBNN: Rethinking Non-linearity for 1-bit CNNs and Going Beyond
• AsymmNet: Towards ultralight convolution neural networks using asymmetrical bottlenecks
• Gradient-Based Interpretability Methods and Binarized Neural Networks