Scribd
Upload
5 views

vgg

The document outlines the implementation of the VGG architecture using a function to create VGG blocks with specified convolutional layers and output channels. It details the construction of the VGG-11 model, including the architecture and memory usage for a single input. Additionally, it describes the training process with a smaller network and provides training loop parameters and results.

Uploaded by

wen zhou
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
5 views

vgg

The document outlines the implementation of the VGG architecture using a function to create VGG blocks with specified convolutional layers and output channels. It details the construction of the VGG-11 model, including the architecture and memory usage for a single input. Additionally, it describes the training process with a smaller network and provides training loop parameters and results.

Uploaded by

wen zhou
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

2/27/2019 vgg slides

Networks Using Blocks (VGG)


We use the vgg_block function to implement this basic VGG block. This function takes
the number of convolutional layers num_convs and the number of output channels
num_channels as input.

In [1]: import d2l


from mxnet import gluon, init, nd
from mxnet.gluon import nn

def vgg_block(num_convs, num_channels):


blk = nn.Sequential()
for _ in range(num_convs):
blk.add(nn.Conv2D(num_channels, kernel_size=3,
padding=1, activation='relu'))
blk.add(nn.MaxPool2D(pool_size=2, strides=2))
return blk

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 1/6
2/27/2019 vgg slides

VGG Architecture

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 2/6
2/27/2019 vgg slides

In [2]: conv_arch = ((1, 64), (1, 128), (2, 256), (2, 512), (2, 512))

Now, we will implement VGG-11. This is a simple matter of executing a for loop over
conv_arch.

In [3]: def vgg(conv_arch):


net = nn.Sequential()
# The convolutional layer part.
for (num_convs, num_channels) in conv_arch:
net.add(vgg_block(num_convs, num_channels))
# The fully connected layer part.
net.add(nn.Dense(4096, activation='relu'), nn.Dropout(0.5),
nn.Dense(4096, activation='relu'), nn.Dropout(0.5),
nn.Dense(10))
return net

net = vgg(conv_arch)

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 3/6
2/27/2019 vgg slides

Memory usage for single input

Next, we will construct a single-channel data example with a height and width of 224 to
observe the output shape of each layer.

In [4]: net.initialize()
X = nd.random.uniform(shape=(1, 1, 224, 224))
for blk in net:
X = blk(X)
print(blk.name, 'output shape:\t', X.shape)

sequential1 output shape: (1, 64, 112, 112)


sequential2 output shape: (1, 128, 56, 56)
sequential3 output shape: (1, 256, 28, 28)
sequential4 output shape: (1, 512, 14, 14)
sequential5 output shape: (1, 512, 7, 7)
dense0 output shape: (1, 4096)
dropout0 output shape: (1, 4096)
dense1 output shape: (1, 4096)
dropout1 output shape: (1, 4096)
dense2 output shape: (1, 10)

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 4/6
2/27/2019 vgg slides

Model Training

Since VGG-11 is more complicated than AlexNet let's use a smaller network.,

In [5]: ratio = 4
small_conv_arch = [(pair[0], pair[1] // ratio) for pair in conv_arch]
net = vgg(small_conv_arch)

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 5/6
2/27/2019 vgg slides

Training Loop

In [6]: lr, num_epochs, batch_size, ctx = 0.05, 5, 128, d2l.try_gpu()


net.initialize(ctx=ctx, init=init.Xavier())
trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': lr})
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size, resize=224)
d2l.train_ch5(net, train_iter, test_iter, batch_size, trainer, ctx, num_epochs)

training on gpu(0)
epoch 1, loss 0.9533, train acc 0.654, test acc 0.852, time 37.7 sec
epoch 2, loss 0.4086, train acc 0.850, test acc 0.886, time 35.8 sec
epoch 3, loss 0.3319, train acc 0.878, test acc 0.900, time 35.9 sec
epoch 4, loss 0.2915, train acc 0.894, test acc 0.904, time 35.9 sec
epoch 5, loss 0.2605, train acc 0.905, test acc 0.911, time 35.8 sec

http://127.0.0.1:8000/vgg.slides.html?print-pdf#/ 6/6

You might also like