This article introduces the pruning principle based on L1 regularization, and uses the VGG network for practical illustrations. The whole process of the model will be introduced in detail from zero 训练、稀疏化、剪枝、finetune
, and detailed source code and instructions will be provided, which will help to master the pruning proficiently, and the pruning of yolov8 will also be introduced later.
论文: Learning Efficient Convolutional Networks through Network Slimming
The author
VGG、DenseNet、ResNet
pruned the model and found50%
that the accuracy of the model is better than that without pruning (BaseLine)更高
, and the accuracy of the BaseLine model can be maintained after more than 60% pruning.
The arrangement of this series of blog posts is as follows:
1 Prerequisites Prerequisite knowledge
- CIFAR10 dataset
- Introduction
- dataset, data loader
- VGG Network
- Introduction to the paper
- model structure
- Neural Network Model Construction
- Batch Normalize
- Introduction to the paper
- Implementation