MindMap Gallery convolutional neural network
The detailed architecture of the convolutional neural network, the bottleneck of the fully connected network, is that as the data size of the image becomes larger, the hidden layer network will be very complicated, come and take a look!
Edited at 2023-07-27 22:50:42This infographic, created using EdrawMax, outlines the pivotal moments in African American history from 1619 to the present. It highlights significant events such as emancipation, key civil rights legislation, and notable achievements that have shaped the social and political landscape. The timeline serves as a visual representation of the struggle for equality and justice, emphasizing the resilience and contributions of African Americans throughout history.
This infographic, designed with EdrawMax, presents a detailed timeline of the evolution of voting rights and citizenship in the U.S. from 1870 to the present. It highlights key legislative milestones, court decisions, and societal changes that have expanded or challenged voting access. The timeline underscores the ongoing struggle for equality and the continuous efforts to secure voting rights for all citizens, reflecting the dynamic nature of democracy in America.
This infographic, created using EdrawMax, highlights the rich cultural heritage and outstanding contributions of African Americans. It covers key areas such as STEM innovations, literature and thought, global influence of music and arts, and historical preservation. The document showcases influential figures and institutions that have played pivotal roles in shaping science, medicine, literature, and public memory, underscoring the integral role of African American contributions to society.
This infographic, created using EdrawMax, outlines the pivotal moments in African American history from 1619 to the present. It highlights significant events such as emancipation, key civil rights legislation, and notable achievements that have shaped the social and political landscape. The timeline serves as a visual representation of the struggle for equality and justice, emphasizing the resilience and contributions of African Americans throughout history.
This infographic, designed with EdrawMax, presents a detailed timeline of the evolution of voting rights and citizenship in the U.S. from 1870 to the present. It highlights key legislative milestones, court decisions, and societal changes that have expanded or challenged voting access. The timeline underscores the ongoing struggle for equality and the continuous efforts to secure voting rights for all citizens, reflecting the dynamic nature of democracy in America.
This infographic, created using EdrawMax, highlights the rich cultural heritage and outstanding contributions of African Americans. It covers key areas such as STEM innovations, literature and thought, global influence of music and arts, and historical preservation. The document showcases influential figures and institutions that have played pivotal roles in shaping science, medicine, literature, and public memory, underscoring the integral role of African American contributions to society.
convolutional neural network
Fully Connected Network Bottleneck
As the data size of the image increases, the hidden layer network will become very complex.
convolutional neural network
layer structure
convolution layer
Convolution kernel
Now for the convolution kernel group
wxya
The depth of the feature response map group is equal to the number of convolution kernels
Different feature response maps reflect the response effect of the input image to different convolution kernels.
The values at different positions on the same feature response map represent the response results of different positions on the input image to the same convolution kernel.
Convolution step size
Perform convolution operations at specified intervals
border padding
effect
Keep input and output sizes consistent
The most common is zero padding
activation layer
Pooling layer
Perform each feature response map independently, reduce the width kernel height of each map in the map group, reduce the number of parameters of subsequent convolutional layers, increase the receptive field, and thereby control overfitting.
operate
Pooling a certain area of the feature response map means specifying a value in the area to represent the entire area.
Common types
max pooling
Equivalent to non-maximizing suppression
average pooling
hyperparameters
pooling window
Pooling step size
Fully connected layer
loss function
cross entropy loss
optimization
SGD
SGD with momentum
ADAM
image enhancement
question
Overfitting occurs because there are too few learning samples, resulting in the inability to train a model that can generalize to new data.
Generate more training data from existing training samples by augmenting the samples with a variety of random transformations that produce credible images
Purpose
Allow the model to observe more content of the data, thereby having better generalization capabilities
sample enhancement
Random scaling and cutout
Training stage: Randomly deduct at different scales and different areas
Testing phase: Withdraw according to a predefined method
color dither
Translation, rotation, distortion, cropping, etc.