Convolutional Neural Networks (CNNs) have made great strides in recent years and are a dazzling pearl in deep learning. CNN can not only be used to classify images, but also has a wide range of applications in image segmentation (object detection) tasks. CNNs have become the gold standard for image classification and are constantly evolving and improving.
Dr. Xin Liu summed up the evolution history of CNN, as shown in the following figure:
The starting point of CNN is the neurocognitive machine model. At this time, the convolution structure has appeared. The classic LeNet was born in 1998. However, the edge of CNN began to be overshadowed by models such as SVM. With the proposal of ReLU and dropout, and the historical opportunities brought by GPU and big data, CNN ushered in a historical breakthrough in 2012: AlexNet. In the following years, CNN showed explosive development, and various CNN models emerged.
The main evolution directions of CNN are as follows:
1. Deepening the network structure
2. Strengthening the convolution function
3. From classification to detection
4. New functional modules
The following figure is a comparison chart of several classic CNN models (AlexNet, VGG, NIN, GoogLeNet, ResNet). It can be seen that the network level is getting deeper and deeper, the structure is getting more and more complex, and of course the model effect is getting better and better:
this blog passes a The series of "Dahua Deep Learning" articles comprehensively and detailedly introduced the milestone achievements of various stages in the evolutionary history of CNN.
1. Xiaobai talks about convolution: Dahua Convolutional Neural Network (CNN)
2. First attempt at convolution: Dahua CNN classic model LeNet 3. Historical breakthrough: Dahua CNN classic model AlexNet 4. Deeper network: Dahua CNN classic model VGGNet 5. Enhanced convolution function: Dahua CNN classic model GoogLeNet 6. From classification to target detection: Dahua target detection model (R-CNN, Fast R-CNN, Faster R-CNN) 7. History of network depth: Dahua depth residual Network (ResNet)
The best way to really get a deep understanding of learning these classic CNN models is to peruse the corresponding papers. Welcome to the public account "Big Data and Artificial Intelligence Lab" (BigdataAILab), which collects and organizes classic papers on various CNN models
Follow my official account "Big Data and Artificial Intelligence Lab" (BigdataAILab), and then reply to the keyword " paper " to read the content of classic papers online .
