Neural-Network-Visualization（VGG-16 image classification based on MNIST dataset）

Neural network visualization (loss-acc graph, mean gradient graph, feature map, convolution kernel, parameters)

1. VGG-16 architecture

Here we use the PlotNeuralNet drawing tool to draw our net's structure. https://github.com/HarisIqbal88/PlotNeuralNet

2. Parameters visualization

Here we use the torchsummary library to summarize our model parameters and the output sizes of each layer.

3. Loss - acc curve graph

The VGG-16 network model was trained ten epochs in total. After each epoch of training, the trained model was put on the verification set to calculate the accuracy.

4. Average Gradient visualization

We visualize the average gradient of each layer in each training epoch, excluding the activation function.(Show part)

5. Convolutional layer visualization

VGG-16 has a total of 5 convolution layers (Layer1-Layer5), and the number of convolution kernel is 64, 128, 256, 512, 512, and we will visualize each convolution kernel.(Show part)

6. Feature map visualization

VGG-16 has a total of 13 convolution layers, and we get the feature map after each convolution layer.