Shulou(Shulou.com)06/01 Report--

What is the PyTorch implementation of convolution neural network on 9 topics? I believe many inexperienced people don't know what to do about it. Therefore, this paper summarizes the causes and solutions of the problem. Through this article, I hope you can solve this problem.

For the implementation of various convolution neural network models, here will introduce their PyTorch implementation, very useful!

1. Typical Network (Classical network)

Typical convolution neural networks include: AlexNet, VGG, ResNet, InceptionV1, InceptionV2, InceptionV3, InceptionV4, Inception-ResNet.

Take the AlexNet network as an example. AlexNet was designed by Hinton, the winner of the 2012 ImageNet competition, and his student Alex Krizhevsky. AlexNet contains several relatively new technical points, and Trick such as ReLU, Dropout and LRN are successfully applied in CNN for the first time. At the same time, AlexNet also uses GPU for operation acceleration.

The PyTorch implementation of AlexNet network structure is as follows:

Import torchimport torch.nn as nndef Conv3x3BNReLU (in_channels,out_channels,stride,padding=1): return nn.Sequential (nn.Conv2d (in_channels=in_channels, out_channels=out_channels, kernel_size=3, stride=stride, padding=1), nn.BatchNorm2d (out_channels), nn.ReLU6 (inplace=True) def Conv1x1BNReLU (in_channels,out_channels): return nn.Sequential (nn.Conv2d (in_channels=in_channels) Out_channels=out_channels, kernel_size=1, stride=1, padding=0), nn.BatchNorm2d (out_channels), nn.ReLU6 (inplace=True)) def ConvBNReLU (in_channels,out_channels,kernel_size,stride,padding=1): return nn.Sequential (in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride, padding=padding), nn.BatchNorm2d (out_channels) Nn.ReLU6 (inplace=True)) def ConvBN (in_channels,out_channels,kernel_size,stride,padding=1): return nn.Sequential (nn.Conv2d (in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride, padding=padding), nn.BatchNorm2d (out_channels)) class ResidualBlock (nn.Module): def _ init__ (self, in_channels,out_channels): super (ResidualBlock) Self). _ init__ () mid_channels = out_channels//2 self.bottleneck = nn.Sequential (ConvBNReLU (in_channels=in_channels, out_channels=mid_channels, kernel_size=1, stride=1), ConvBNReLU (in_channels=mid_channels, out_channels=mid_channels, kernel_size=3, stride=1, padding=1), ConvBNReLU (in_channels=mid_channels, out_channels=out_channels, kernel_size=1, stride=1) ) self.shortcut = ConvBNReLU (in_channels=in_channels, out_channels=out_channels, kernel_size=1, stride=1) def forward (self, x): out = self.bottleneck (x) return out+self.shortcut (x) 2. Lightweight Network (Lightweight)

Lightweight networks include: GhostNet, MobileNets, MobileNetV2, MobileNetV3, ShuffleNet, ShuffleNet V2, SqueezeNet Xception MixNet GhostNet.

Take GhostNet as an example, the same accuracy, speed and computational complexity are less than the previous SOTA algorithm. The core of GhostNet is the Ghost module. Compared with the ordinary convolution neural network, the total number of parameters and computational complexity are reduced without changing the size of the output feature graph, and plug and play.

The PyTorch implementation of GhostNet network structure is as follows:

Https://github.com/shanglianlm0525/PyTorch-Networks/blob/master/Lightweight/GhostNet.py

3. Target detection network (ObjectDetection)

The target detection network includes: SSD, YOLO, YOLOv2, YOLOv3, FCOS, FPN, RetinaNet Objects as Points, FSAF, CenterNet FoveaBox.

Taking YOLO series as an example, YOLO (You Only Look Once) is an object recognition and location algorithm based on depth neural network. Its biggest feature is that it runs fast and can be used in real-time systems. At present, there are many YOLOv3 applications.

The PyTorch implementation of YOLOV3 network structure is as follows:

Https://github.com/shanglianlm0525/PyTorch-Networks/blob/master/ObjectDetection/YOLOv3.py

4. Semantic Segmentation Network (SemanticSegmentation)

Semantic segmentation network includes: FCN, Fast-SCNN, LEDNet, LRNNet, FisheyeMODNet.

Take FCN as an example. FCN was born in the pioneer of semantic segmentation model in 2014. Its main contribution is to promote the use of end-to-end convolution neural network in semantic segmentation and use deconvolution for up-sampling. The FCN model is very simple, it is all composed of convolution, so it is called full convolution network, and because of the special form of full convolution, it can accept input of any size.

The PyTorch implementation of FCN network structure is as follows:

Https://github.com/shanglianlm0525/PyTorch-Networks/blob/master/SemanticSegmentation/FCN.py

5. Instance segmentation network (InstanceSegmentation)

Example segmentation network includes: PolarMask.

6. Face Detection and recognition Network (commit VarGFaceNet)

Face detection and recognition network includes: FaceBoxes, LFFD, VarGFaceNet.

7. Human posture recognition Network (HumanPoseEstimation)

Human posture recognition network includes: Stacked Hourglass, Networks Simple Baselines, LPN.

8. Attention mechanism network

Attention mechanism network includes: SE Net, scSE, NL Net, GCNet, CBAM.

9. Portrait Segmentation Network (PortraitSegmentation)

The portrait segmentation network includes: SINet.

To sum up, the GitHub open source project shows nine main types of convolution neural networks in recent years, including dozens of specific network structures. Each of these network structures has its own PyTorch implementation. It's still pretty good.

After reading the above, have you mastered the PyTorch implementation of the convolution neural network of 9 major topics? If you want to learn more skills or want to know more about it, you are welcome to follow the industry information channel, thank you for reading!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.