Matsu
deep learning with pytorch_manning_ch.02 본문
deep learning/deep learning with pytorch_manning
deep learning with pytorch_manning_ch.02
Matsu 2021. 6. 12. 20:41
2. Pretrained networks¶
We are going to learn how to use the work of the best researchers in the field
by downloading and running very interesting models that have already been traind on open, large-scale datsets
We can think of a pretrained neural network as similar to a program that takes inputs and generates outputs.
In this chapter, we will explore three popular pretrained models:
a model that can label an image according to its content, another that can fabricate a new image from a real image,
and a model that can describe the content of an image using proper English sentences.
2.1 A pretrained network that recognizes the subject of an image¶
We are going to end up being able to take our own images and feed them into our pretrained model.
This will result in a list of predicted labels for that image,
which we can then examine to see what the model thinks our image is.
The input image will first be preprocessed into an instance of the multidimensional array class torch.Tensor.
It is an RGB image with height and width, so this tensor will have three dimensions:
the three color channels, and two spatial image dimensions of a specific size.
Our model will take that processed input image and pass it into the pretrained network to obtain scores for each class.
Before we can do all that, we'll need to get the network itself, take a peek under the hood to see how it's structured,
and learn about how to prepare our data beore the model can use it.
2.1.1 Obtaining a pretrained network for image recognition¶
For now, let's load up and run two networks:first AlexNet, one of the early breakthrough networks for image recognition;
and then a residual network, ResNet for short,
which won the ImageNet classification, detection, and localization competitions, among others, in 2015.
In [2]:
from torchvision import models
In [3]:
dir(models)
Out[3]:
['AlexNet',
'DenseNet',
'GoogLeNet',
'GoogLeNetOutputs',
'Inception3',
'InceptionOutputs',
'MNASNet',
'MobileNetV2',
'MobileNetV3',
'ResNet',
'ShuffleNetV2',
'SqueezeNet',
'VGG',
'_GoogLeNetOutputs',
'_InceptionOutputs',
'__builtins__',
'__cached__',
'__doc__',
'__file__',
'__loader__',
'__name__',
'__package__',
'__path__',
'__spec__',
'_utils',
'alexnet',
'densenet',
'densenet121',
'densenet161',
'densenet169',
'densenet201',
'detection',
'googlenet',
'inception',
'inception_v3',
'mnasnet',
'mnasnet0_5',
'mnasnet0_75',
'mnasnet1_0',
'mnasnet1_3',
'mobilenet',
'mobilenet_v2',
'mobilenet_v3_large',
'mobilenet_v3_small',
'mobilenetv2',
'mobilenetv3',
'quantization',
'resnet',
'resnet101',
'resnet152',
'resnet18',
'resnet34',
'resnet50',
'resnext101_32x8d',
'resnext50_32x4d',
'segmentation',
'shufflenet_v2_x0_5',
'shufflenet_v2_x1_0',
'shufflenet_v2_x1_5',
'shufflenet_v2_x2_0',
'shufflenetv2',
'squeezenet',
'squeezenet1_0',
'squeezenet1_1',
'utils',
'vgg',
'vgg11',
'vgg11_bn',
'vgg13',
'vgg13_bn',
'vgg16',
'vgg16_bn',
'vgg19',
'vgg19_bn',
'video',
'wide_resnet101_2',
'wide_resnet50_2']The capitalized names refer to Python classes that implement a number of popular models.
We'll now turn our attention to AlexNet.
2.1.2 AlexNet¶
The AlexNet architecture won the 2012 ILSVRC.
Alexnet is a rather small network, but it's perfect for taking a first peek at a neural network that does something and learning how to run a pretrained version of it on a new image.
In order to run the AlexNet architecture on an input image,
we can create an instance of the AlexNet class.
In [4]:
alexnet = models.AlexNet()
2.1.3 ResNet¶
Using the resnet101 function, we'll now instantiate a 101-layer convolutional neural network.
In [5]:
resnet = models.resnet101(pretrained=True)
Downloading: "https://download.pytorch.org/models/resnet101-5d3b4d8f.pth" to C:\Users\CJW/.cache\torch\hub\checkpoints\resnet101-5d3b4d8f.pth
2.1.4 Ready, set, almost run¶
We'll take a peek at what a resnet101 looks like.
In [7]:
resnet
Out[7]:
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(6): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(7): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(8): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(9): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(10): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(11): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(12): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(13): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(14): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(15): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(16): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(17): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(18): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(19): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(20): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(21): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(22): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)
)What we are seeing here is modules, one per line.
Note that they have nothing in common with Python modules.
They are individual operations, the building blocks of a neural network.
They are also called layers in other deep learning frameworks.
In [8]:
from torchvision import transforms
preprocess = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(
mean = [0.485, 0.456, 0.406],
std = [0.229, 0.224, 0.225])
])
In this case, we defined a preprocess function that will scale the input image to 256 X 256,
crop the image to 224 X 224 around the center, transform it to a tensor and normalize its RGB components
so that they have defined means and standard deviations.
In [11]:
from PIL import Image
In [14]:
img = Image.open('bobby.jpg')
img
Out[14]:
In [15]:
img_t = preprocess(img)
In [16]:
import torch
batch_t = torch.unsqueeze(img_t, 0)
2.1.5 Run!¶
The process of running a trained model on new data is called inference in deep learning circles.
In order to inference, we need to put the network in eval mode:
In [17]:
resnet.eval()
Out[17]:
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(6): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(7): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(8): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(9): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(10): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(11): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(12): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(13): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(14): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(15): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(16): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(17): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(18): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(19): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(20): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(21): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(22): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)
)In [18]:
out = resnet(batch_t)
out
Out[18]:
tensor([[-3.4997e+00, -1.6490e+00, -2.4391e+00, -3.2243e+00, -3.2465e+00,
-1.3218e+00, -2.0395e+00, -2.5405e+00, -1.3043e+00, -2.8827e+00,
-1.6696e+00, -1.2838e+00, -2.6184e+00, -2.9750e+00, -2.4380e+00,
-2.8256e+00, -3.3083e+00, -7.9667e-01, -6.7075e-01, -1.2162e+00,
-3.0311e+00, -3.9593e+00, -2.2631e+00, -1.0843e+00, -9.7916e-01,
-1.0742e+00, -3.0908e+00, -2.4751e+00, -2.2153e+00, -3.1932e+00,
-3.2964e+00, -1.8507e+00, -2.0642e+00, -2.1202e+00, -1.8665e+00,
-3.2375e+00, -1.1210e+00, -1.1321e+00, -1.1657e+00, -9.0362e-01,
-4.5209e-01, -1.4986e+00, 1.4366e+00, 1.2994e-01, -1.8379e+00,
-1.4815e+00, 9.7278e-01, -9.3662e-01, -3.0276e+00, -2.7341e+00,
-2.5960e+00, -2.0591e+00, -1.8170e+00, -1.9437e+00, -1.7875e+00,
-1.3029e+00, -4.5200e-01, -2.0560e+00, -3.2882e+00, -4.7583e-01,
-3.6261e-01, -1.1650e+00, -7.3942e-01, -1.4489e+00, -1.5039e+00,
-2.1096e+00, -1.7095e+00, -4.1315e-01, -1.9146e+00, -1.5095e+00,
-1.2012e+00, -1.3111e+00, -1.0662e+00, 1.2968e-01, -3.8648e-01,
-2.4670e-01, -8.7028e-01, 5.9567e-01, -8.8329e-01, 1.4535e+00,
-2.6384e+00, -3.6590e+00, -2.3378e-01, -4.9410e-02, -2.2602e+00,
-2.3605e+00, -1.4038e+00, 2.4098e-01, -1.0531e+00, -2.9106e+00,
-2.5321e+00, -2.2253e+00, 4.4062e-01, -1.3269e+00, -2.0447e-02,
-2.8696e+00, -5.5842e-01, -1.3959e+00, -2.9274e+00, -1.9038e+00,
-4.2433e+00, -2.9701e+00, -2.0552e+00, -2.4290e+00, -2.7691e+00,
-4.0207e+00, -3.6511e+00, -4.8903e-01, -9.8899e-01, -1.8995e+00,
-3.5459e+00, -1.4606e+00, 1.1054e+00, -6.9230e-01, -7.4877e-01,
-2.1745e+00, -2.2471e+00, -5.3973e-01, -1.5121e+00, -2.5998e+00,
-3.9889e+00, -9.7558e-01, -5.5630e-01, -8.6530e-01, -1.2110e+00,
-7.3269e-01, -1.0847e+00, -2.0944e+00, -4.0080e+00, -3.3433e-01,
-2.6919e+00, -2.9908e+00, -1.8793e+00, -1.8126e+00, -1.2953e+00,
-2.2093e+00, -2.0246e+00, -3.1878e+00, -3.2739e+00, -2.7969e+00,
-4.1270e-01, -3.7697e+00, -2.4744e+00, -2.6133e+00, -2.7514e+00,
-2.6512e+00, -3.2751e+00, -4.3150e+00, -4.2214e+00, -3.5920e+00,
-1.2186e+00, 2.3577e+00, 1.9648e+00, 2.2958e+00, 5.0050e+00,
1.0229e+00, 2.9914e+00, 8.8864e-01, 2.2502e+00, 6.1697e+00,
4.2833e+00, 4.5931e+00, 7.4687e+00, 4.3448e+00, 4.8805e+00,
5.8076e+00, 3.9792e+00, 3.5555e+00, 9.5253e+00, 1.1172e+00,
3.3158e+00, 1.9011e+00, -5.1524e-01, 1.3985e+00, 1.8150e+00,
5.6135e+00, 5.7316e+00, 2.1489e+00, 2.4986e+00, 5.5508e-01,
1.8658e+00, 1.4871e+00, 3.4364e+00, 4.4125e-01, 4.2860e+00,
4.3832e+00, 1.3132e+00, 1.6830e+00, 1.1566e+00, 2.1403e+00,
9.1246e-01, 3.0220e+00, 3.5398e+00, 2.8193e+00, 1.9450e+00,
8.4406e-02, 2.0383e+00, 2.7105e+00, 1.0817e+00, 1.9388e+00,
3.5678e+00, 2.3330e+00, 2.5193e+00, 1.3185e+00, 3.6052e+00,
7.7896e+00, 4.4019e+00, 1.5813e+01, 1.2201e+01, 5.1687e+00,
1.9371e+00, 5.5298e+00, 6.2411e+00, 7.5443e+00, 5.9197e+00,
7.0106e+00, 5.7829e+00, 2.6940e+00, 5.3224e+00, 9.9329e+00,
6.4925e+00, 2.4706e+00, 5.6329e+00, 3.4480e+00, 2.0878e+00,
1.2667e+00, 2.5974e+00, 5.6932e+00, 7.1867e-01, 1.0250e+00,
4.8133e+00, 6.0352e+00, 3.9970e+00, -1.7816e+00, 3.6096e+00,
2.8417e+00, 2.8397e+00, 1.2274e+00, 4.2220e+00, 4.0891e+00,
3.3476e+00, 2.3510e+00, 2.6175e-01, 8.3823e-01, 5.1252e+00,
-7.6088e-01, 1.9809e+00, 2.8381e+00, 2.4766e+00, 1.8687e+00,
5.8744e-01, 3.4939e+00, -5.0017e-01, 1.8486e+00, 4.4583e-01,
2.4148e+00, 3.4823e+00, 5.1915e+00, 1.8092e+00, 3.9559e+00,
4.3692e+00, 3.1688e-01, -4.8611e-01, 5.6470e+00, 3.6534e+00,
2.4436e+00, 3.6937e+00, 3.3958e+00, -9.6209e-01, -2.2416e-01,
1.2344e-01, 1.1682e+00, -5.7349e-02, 4.4999e+00, -1.8748e+00,
-2.6818e+00, -1.6847e+00, -6.7984e-01, -1.5588e+00, -2.1798e+00,
-1.0431e+00, 1.5370e+00, 7.6586e-01, -1.7835e+00, 6.3795e-01,
8.8728e-01, -2.1297e+00, -2.1428e+00, -1.5652e+00, -4.8034e+00,
-1.0672e+00, -9.5017e-01, -2.0649e+00, 2.0019e+00, -6.4177e-01,
-2.5378e+00, -1.5008e+00, -3.1975e+00, -1.9333e+00, -5.8521e-01,
-1.0363e+00, 1.3329e+00, -1.4872e+00, -2.0102e+00, 1.3183e+00,
4.2174e-01, -3.5856e-01, 7.4732e-01, -2.7741e-01, -7.5281e-01,
-5.5096e-01, 1.0833e+00, 1.3690e-01, 5.3098e-02, 1.2991e+00,
-2.0118e+00, -1.7402e+00, 1.1788e+00, -4.2505e-01, 3.0964e-01,
-1.4589e+00, -2.6225e+00, -2.4953e+00, -2.1100e-01, -2.3407e+00,
-1.5377e+00, -2.4802e+00, -8.0411e-01, 2.1876e-01, -2.8007e+00,
-1.9041e+00, -9.3844e-01, -3.4383e-01, -4.4378e-01, -7.5068e-01,
-2.5060e+00, -1.9128e+00, -2.1313e+00, -5.4917e-01, 1.7572e-01,
-2.0437e+00, -1.7883e+00, -2.4830e+00, -3.8768e+00, -4.4253e+00,
-2.1819e+00, -2.0485e+00, -3.7339e+00, -1.9185e+00, -3.4523e+00,
-2.0103e+00, -2.2686e+00, -7.9842e-01, -5.0350e-01, -2.9496e+00,
-1.6959e+00, -2.9049e-01, -2.8050e+00, -1.8296e+00, 9.0892e-02,
-2.1747e+00, -1.5452e+00, -1.2558e+00, -9.5389e-01, -2.5265e+00,
-1.3665e+00, -3.8682e+00, -4.3242e+00, -1.5148e+00, -1.9113e+00,
-2.9872e+00, -2.9385e+00, -4.8581e+00, -2.4930e+00, -1.3708e+00,
-2.8970e+00, -1.2814e+00, -5.6305e-01, -2.8289e+00, -1.8666e+00,
-2.3429e+00, -3.4331e+00, -2.5701e+00, -3.8991e+00, -4.4677e+00,
-3.5612e+00, -9.3197e-01, -2.3963e+00, -2.0673e+00, -1.8581e+00,
-3.2819e+00, -3.5219e+00, -2.8122e+00, -1.9661e+00, -3.7761e+00,
-1.1609e+00, -2.7743e+00, 2.5781e-01, -1.7330e-01, -3.0291e-01,
1.6415e-01, 1.7316e+00, 1.8974e+00, -3.1201e+00, -3.7529e+00,
-2.6532e+00, -6.6325e-01, -8.5835e-01, -1.3166e+00, 4.9872e-01,
-2.3256e+00, 4.7837e+00, 1.4947e+00, -1.6190e+00, 3.3725e+00,
3.0711e+00, 1.5550e+00, 7.3322e-01, 1.2925e+00, 1.6615e+00,
9.5885e-02, 1.3348e+00, 2.9194e-02, 9.8533e-01, -5.1703e-01,
-1.3542e+00, 5.3266e-01, 2.7115e-01, 2.8676e+00, 2.2308e+00,
1.8468e+00, 1.3861e+00, 1.1031e+00, 2.3657e+00, 6.4129e+00,
1.8896e+00, 5.8361e-01, -1.5098e+00, -1.8949e+00, -2.6363e+00,
1.5000e+00, -5.3598e-01, -6.9935e-02, 2.0223e+00, -1.7648e+00,
4.1311e-01, 1.6696e+00, 7.3901e-01, 2.1839e+00, -1.4280e+00,
-2.4708e+00, 4.8153e-01, 3.3817e+00, -2.9651e-01, 6.8331e-02,
1.1068e+00, -1.3973e+00, 1.5518e+00, -2.2464e+00, -3.6723e-01,
-1.5056e+00, -2.1660e+00, 2.7747e+00, 2.5229e+00, 1.8796e-01,
4.6500e-01, -1.9905e+00, -1.6224e+00, -7.7251e-01, -1.1583e+00,
-3.0849e-01, -1.4386e+00, 2.2714e+00, -4.5280e-01, -8.7969e-01,
4.1363e-01, -1.4095e+00, -2.8906e-01, 3.4809e+00, 4.3324e-01,
4.2387e-01, -1.1204e+00, 1.9781e-01, -2.6528e+00, -2.8554e+00,
-6.6075e-01, -3.1026e-01, 7.8641e-01, 1.8965e-01, -4.5897e-01,
-5.5900e-01, -2.8286e-01, 1.6559e+00, 7.1157e-02, -4.2351e-01,
1.1153e+00, 2.2649e+00, -1.0259e+00, 9.7924e-01, 1.0194e+00,
-2.6048e+00, -5.5008e-01, 2.1353e+00, -9.3641e-01, 4.3959e-02,
-3.9618e-01, 1.5395e+00, -1.0598e+00, 1.8685e+00, 1.4945e+00,
-4.4445e+00, -1.7148e-01, -1.6854e-01, -7.5243e-01, 1.8241e+00,
3.3237e+00, 7.4655e-01, -7.6161e-01, 1.6335e+00, 1.9438e+00,
1.2983e+00, -1.5951e+00, 2.2410e+00, 9.3765e-01, -6.2485e-01,
-1.3969e+00, 1.6010e+00, 2.9322e-02, 8.8535e-01, 9.6981e-01,
5.0879e-01, 3.9120e-01, 1.0257e+00, 6.5950e-01, 3.3687e+00,
-1.0271e+00, -2.9353e+00, -7.5909e-01, -7.9742e-01, 7.0959e+00,
-3.3062e+00, 1.3640e+00, 1.7516e+00, 3.7723e+00, -1.1675e+00,
5.5042e-01, -1.8410e+00, -1.3717e+00, 4.1994e-01, 2.6614e+00,
1.4409e+00, 8.2754e-01, 4.9228e+00, 1.5345e+00, -3.6179e+00,
-2.2328e+00, -1.3824e+00, -8.8731e-01, 2.2056e+00, 2.0819e+00,
-9.1448e-01, -1.7883e+00, -1.4364e+00, 1.0453e+00, -7.3781e-01,
-1.4398e+00, 4.6885e-01, 8.8505e-01, 6.7727e-01, -1.1915e+00,
-1.4081e+00, -8.5758e-01, 1.1674e+00, -1.8481e+00, 2.5000e+00,
3.4197e-01, -1.7550e+00, -8.9500e-01, 2.9997e+00, 2.2094e+00,
-8.3813e-01, -2.0126e+00, 9.3872e-01, -2.1452e+00, 3.8367e-02,
2.4887e-01, -1.0520e-01, -6.5314e-01, 8.0483e-01, 4.0207e+00,
-1.2178e+00, 2.6536e+00, -1.8085e+00, 7.5238e-01, 5.0293e-01,
1.1506e-02, -7.7166e-01, 4.5107e-01, 1.3278e+00, 8.2390e-01,
-6.1212e-01, 3.0906e-02, 2.6237e-01, -2.1468e+00, -1.4071e+00,
-9.9406e-02, -4.3217e-01, 4.8178e-01, 1.2984e+00, 7.5574e-01,
8.5152e-02, -1.3871e-01, -8.5174e-01, 8.1902e-01, 7.0772e-01,
2.1356e+00, 7.1527e-01, -1.9824e-01, 2.4378e+00, 1.8625e+00,
1.0683e+00, 1.0118e+00, 1.4432e+00, -1.9515e-02, -1.6767e-01,
-1.6930e+00, -1.2930e+00, -1.3439e+00, -3.4781e+00, 5.9320e-01,
1.0796e+00, -1.8917e-02, -3.9900e-01, -4.6973e-01, -3.7134e-01,
-1.5430e+00, 3.1286e-01, -1.8760e-01, 3.2949e+00, 1.8577e+00,
-1.5585e+00, 2.9257e+00, -7.3537e-01, 6.2360e-02, -4.3240e-02,
-1.0193e+00, 1.1614e+00, 6.1005e-01, 1.4297e-01, -3.4956e+00,
1.3337e+00, 3.8960e-01, -1.2900e+00, -1.2823e+00, -1.6482e+00,
1.9285e+00, 2.7965e-02, -9.3298e-01, 6.6934e-01, 2.9179e+00,
1.7228e+00, 2.1640e-01, -8.5777e-01, -2.0391e+00, 1.6510e+00,
-2.3326e-01, 7.4310e-01, 1.5400e+00, -1.4460e+00, -1.2114e+00,
5.2273e-01, -6.4955e-01, 4.8662e-02, 2.6569e+00, -1.6518e+00,
-3.3896e+00, 2.1350e+00, 9.0318e-01, 1.3904e+00, -3.6350e-01,
2.0223e-01, 1.4554e+00, -1.1211e+00, 1.8399e+00, -1.2923e+00,
-2.1249e+00, -8.6377e-01, -5.1445e-01, -1.2872e+00, 1.1431e+00,
-1.4240e+00, -7.6691e-01, -3.5563e-01, 2.1923e+00, -1.9831e-01,
-1.8438e+00, 9.6279e-02, 1.7458e+00, 1.7894e-01, 8.0167e-01,
4.7613e+00, -7.3425e-01, -1.6478e+00, -8.1854e-01, -1.5260e+00,
3.4332e+00, 4.9594e-01, -3.4153e-01, -5.1844e-01, 3.1711e-01,
1.8911e+00, -1.2033e+00, -2.0005e+00, -1.2113e-01, -7.0711e-01,
-2.5450e+00, -6.7446e-02, 8.0279e-01, -1.8256e+00, 2.9197e-01,
-1.1033e+00, 1.5367e+00, 2.9367e+00, 5.3937e-01, -7.6932e-01,
-9.2092e-01, -1.7367e+00, -1.0012e+00, 9.3510e-01, -4.1548e-02,
-2.3282e+00, 2.2907e+00, 9.3231e-01, 2.0218e+00, -1.7313e+00,
2.6369e+00, 3.8262e+00, 6.9461e-01, -2.5077e-03, -9.9190e-01,
2.1992e+00, 1.4213e+00, 4.3468e-01, -1.3524e+00, -8.5725e-01,
5.7249e-01, -6.5436e-01, 5.3128e-01, 1.1928e+00, 9.9297e-01,
2.8021e+00, -3.0989e+00, 3.5913e+00, -1.7487e-01, -7.0675e-02,
-2.0895e+00, -1.4625e+00, 3.0381e+00, -4.9794e-01, 2.0560e-01,
-1.4915e-01, 2.2701e+00, 3.2218e-01, 9.6720e-01, -1.1802e+00,
3.3260e+00, -1.8322e+00, 3.2782e+00, -1.7115e-01, 1.3883e+00,
6.0959e-01, 4.1543e-01, -7.9037e-01, 3.6867e-01, 1.9525e+00,
4.2260e-01, 2.4431e+00, -1.9236e+00, 1.2816e-02, 2.2662e+00,
-1.6735e+00, -7.1518e-01, 1.3959e+00, -2.5440e-02, 7.7251e-01,
3.0006e+00, -9.7998e-01, -1.3561e+00, 2.3763e-01, 2.3640e+00,
1.9144e+00, 1.8060e+00, 8.6761e-01, 4.2975e+00, 5.3407e-01,
-1.4330e+00, -5.0711e-01, 3.0123e+00, -8.6910e-01, 3.2884e+00,
-6.6958e-01, -2.6305e+00, -3.1957e+00, -2.9512e+00, 1.2077e+00,
5.6051e+00, 2.7645e-01, -3.1766e+00, 2.5924e+00, -3.3277e-01,
1.2678e-01, 2.3057e+00, -1.4350e+00, 3.3605e+00, -3.3686e+00,
-6.4941e-01, 1.1768e+00, -2.6352e+00, 1.6955e+00, -1.4452e+00,
-2.6586e+00, -3.6629e+00, -1.7268e+00, 1.1226e+00, 1.9893e+00,
-1.6089e-01, 1.2605e+00, 1.2145e+00, 2.6831e+00, -1.8129e+00,
-1.8627e+00, 1.5719e+00, -2.0970e+00, -2.3203e+00, 9.8824e-01,
-3.8493e+00, -1.1166e+00, 8.9900e-01, -3.4761e-01, -3.5048e+00,
2.0611e+00, 5.2646e-01, 6.1806e-01, -1.1572e-01, 8.0079e-01,
8.0897e-01, 1.0604e+00, -1.8260e+00, 2.6638e-01, -1.1470e+00,
8.3008e-01, 1.0902e+00, 9.0343e+00, -9.1355e-01, -9.0900e-01,
1.1300e+00, -3.8219e-01, -1.7321e+00, 1.0250e-01, 2.0110e-01,
-1.4002e+00, 8.6910e-01, -4.4545e-01, -7.9011e-01, -3.1371e+00,
-3.6089e-01, 1.9827e-01, -2.1941e+00, 2.8923e+00, 4.7277e-01,
1.4522e+00, -3.4306e+00, 1.4684e+00, -2.5527e+00, -2.9418e+00,
-7.8795e-01, 2.8671e+00, -1.2269e+00, -6.5231e-01, 3.6328e+00,
1.1509e+00, 1.1695e+00, 3.3913e+00, 9.2258e-01, -1.1167e+00,
1.0877e+00, 1.9697e-01, -7.4067e-01, -2.9546e+00, -1.7774e-01,
1.5686e+00, -4.4219e-01, 1.6109e+00, 1.0286e+00, -1.2968e+00,
-1.7997e+00, -4.9459e-01, 5.5377e-01, 2.3711e+00, 5.1920e-01,
-1.3707e+00, -2.3571e+00, -7.6039e-01, 1.3218e+00, 2.9741e+00,
-2.6011e-01, -6.0607e-01, -4.2339e-02, -7.4892e-01, -7.3591e-02,
2.7876e+00, 3.2539e-01, -1.1395e+00, -1.4875e+00, -4.3684e+00,
-3.6234e-01, -3.4596e-01, -2.0751e+00, 3.6143e-01, -1.4748e+00,
-6.3914e-01, -2.0747e+00, -9.1273e-01, -4.0377e-01, -3.3568e-01,
-1.8499e+00, -1.7760e+00, -6.6116e-01, 6.3051e-01, 3.7449e+00,
2.1652e+00, 4.1343e+00, 1.6970e+00, 8.3944e-01, 1.2930e+00,
1.0517e+00, 1.4064e+00, 1.3186e+00, -7.6195e-01, 2.2661e+00,
1.4491e-01, -7.7404e-01, 1.0131e+00, 8.7297e-01, -6.4795e-01,
-3.4213e-01, -1.5297e+00, 2.0686e+00, 2.5062e+00, 2.1082e-01,
7.5025e-01, 1.8743e-02, -1.7822e+00, 1.9568e+00, 4.3805e-01,
1.2335e+00, 7.7087e-01, -1.0329e+00, 1.5594e-01, -3.6626e-01,
6.3101e-02, 3.2490e+00, 6.4807e-01, -1.5173e-01, 9.3229e-01,
-6.0559e-02, -1.1985e+00, 1.4659e-01, 3.9304e-01, 8.8496e-01,
-1.9458e+00, 1.2217e-01, -1.5016e+00, -1.8235e+00, -3.8595e+00,
1.8626e-01, -2.9675e+00, 5.4217e-01, -7.8127e-01, -2.6196e+00,
-4.4958e+00, 5.7119e-01, -1.3398e+00, -3.8117e+00, -7.8630e-01,
-5.6788e-01, -2.5453e+00, 2.4054e+00, 6.5602e-01, 4.7648e-01,
2.8749e+00, -3.7451e+00, 1.5124e+00, -3.2777e+00, -2.4971e+00,
-3.2263e-01, 1.2816e-01, -1.1752e+00, 3.4434e+00, 4.4534e+00]],
grad_fn=<AddmmBackward>)We now need to find out the label of the class that receeived the highest score.
This will tell us what the model saw in the image.
If the label matches how a human would describe the image, that's great! It means everything is working.
If not, either something went wrong during training, or the image is so different from what the model expects
that the model can't process it properly, or there's some other similar issue.
Let's load the file containing the 1,000 labels for the ImageNet dataset classes.
In [26]:
with open ('imagenet_classes.txt') as f:
labels = [line.strip() for line in f.readlines()]
At this point, we need to determine the index corresponding to the maximum score in the out tensor we obtained previously.
We can do that using the max function in PyTorch, which outputs the maximum value in a tensor as well as the indices
where that maximum value occurred.
In [27]:
_, index = torch.max(out, 1)
We can now use the index to access the label.
In [29]:
percentage = torch.nn.functional.softmax(out, dim=1)[0] * 100
labels[index[0]], percentage[index[0]].item()
Out[29]:
('golden retriever', 96.57185363769531)cf. 소프트맥스는 세 개 이상으로 분류하는 다중 클래스 분류에서 사용되는 활성화 함수다.
소프트맥스 함수는 분류될 클래스가 n개라 할 때, n차원의 벡터를 입력받아, 각 클래스에 속할 확률을 추정한다.
We can also find out what the second best, third best, and so on were.
To do this, we can use the sort function, which sorts the values in ascending or descending order
and also provides the indices of the sorted values in the original array.
In [31]:
_, indices = torch.sort(out, descending = True)
[(labels[idx], percentage[idx].item()) for idx in indices[0][:5]]
Out[31]:
[('golden retriever', 96.57185363769531),
('Labrador retriever', 2.6082682609558105),
('cocker spaniel, English cocker spaniel, cocker', 0.2699616849422455),
('redbone', 0.17959021031856537),
('tennis ball', 0.10991999506950378)]2.2 A pretrained model that fakes it until it makes it¶
2.2.1 The GAN game¶
career criminal painter versus art historian.
In the context of deep learning, what we've just described is known as the GAN game,
where two networks, one acting as the painter and the other as the art historian,
compete to outsmart each other at creating and detecting forgeries.
GAN stands for generative adversarial network, where generative means something is being created(in this case, fake masterpieces),
adversarial means the two networks are competing to outsmart the otehr, and well, network is pretty obvious.
Remember that our overarching goal is to produce synthetic examples of a class of images that cannot be recognized as fake.
When mixed in with legitimate examples, a skilled examiner would have trouble determining which ones are real
and which are our forgeries.
Rough picture of what's going on.
The end goal for the generator isi to fool the discriminator into mixing up real and fake images.
The end goal for the discriminator is to find out when it's being tricked,
but it also helps inform the generator about the identifiable mistakes in the generated images.
As training progresses, information flows back from the discriminator, and the generator uses it to improve.
This technique has proven itself able to lead to generators that produce realistic images
from nothing but noise and a conditioning signal, like an attribute or another image.
In other words, a well-trained generator learns a plausible model for generating images that look real even when examined by humans.
In figure, we have a cycleGAN workflow for the task of turning a photo of a dude into a dog, and vice versa.
Note that there are two separate generator networks, as well as two distinct discriminators.
The first generator learns to produce an image conforming to a traget distribution (dogs, in this case)
starting from an image belonging to a different distribution(dude),
so that the discriminator can't tell if the image produced from a dude photo is actually a genuine picutre of a dog or not.
At the same time, the resulting fake dog is sent through a dfferent generator going the other way (dog to dude, in our case),
to be judged by another discriminator on the other side.
Creating such a cyle stabilizes the training processs considerably, which addresses one of the original issues with GANs.
2.2.3 A network that turns horses into zebras¶
We can play with this model right now.
The CycleGAN network has been trained on a dataset of (unrelatd) horse images and zebra images extracted from the ImageNet dataset.
Playing with a pretrained CycleGAN will give us the opportunity to take a step closer and look at how a network-a generator,
in this case- is implemneted. We'll use our old friend ResNet.
In [33]:
# copy from code/p1ch2/3_cyclegan.ipynb
import torch
import torch.nn as nn
class ResNetBlock(nn.Module): # <1>
def __init__(self, dim):
super(ResNetBlock, self).__init__()
self.conv_block = self.build_conv_block(dim)
def build_conv_block(self, dim):
conv_block = []
conv_block += [nn.ReflectionPad2d(1)]
conv_block += [nn.Conv2d(dim, dim, kernel_size=3, padding=0, bias=True),
nn.InstanceNorm2d(dim),
nn.ReLU(True)]
conv_block += [nn.ReflectionPad2d(1)]
conv_block += [nn.Conv2d(dim, dim, kernel_size=3, padding=0, bias=True),
nn.InstanceNorm2d(dim)]
return nn.Sequential(*conv_block)
def forward(self, x):
out = x + self.conv_block(x) # <2>
return out
class ResNetGenerator(nn.Module):
def __init__(self, input_nc=3, output_nc=3, ngf=64, n_blocks=9): # <3>
assert(n_blocks >= 0)
super(ResNetGenerator, self).__init__()
self.input_nc = input_nc
self.output_nc = output_nc
self.ngf = ngf
model = [nn.ReflectionPad2d(3),
nn.Conv2d(input_nc, ngf, kernel_size=7, padding=0, bias=True),
nn.InstanceNorm2d(ngf),
nn.ReLU(True)]
n_downsampling = 2
for i in range(n_downsampling):
mult = 2**i
model += [nn.Conv2d(ngf * mult, ngf * mult * 2, kernel_size=3,
stride=2, padding=1, bias=True),
nn.InstanceNorm2d(ngf * mult * 2),
nn.ReLU(True)]
mult = 2**n_downsampling
for i in range(n_blocks):
model += [ResNetBlock(ngf * mult)]
for i in range(n_downsampling):
mult = 2**(n_downsampling - i)
model += [nn.ConvTranspose2d(ngf * mult, int(ngf * mult / 2),
kernel_size=3, stride=2,
padding=1, output_padding=1,
bias=True),
nn.InstanceNorm2d(int(ngf * mult / 2)),
nn.ReLU(True)]
model += [nn.ReflectionPad2d(3)]
model += [nn.Conv2d(ngf, output_nc, kernel_size=7, padding=0)]
model += [nn.Tanh()]
self.model = nn.Sequential(*model)
def forward(self, input): # <3>
return self.model(input)
In [34]:
netG = ResNetGenerator()
In [37]:
model_path = 'horse2zebra_0.4.0.pth'
model_data = torch.load(model_path)
netG.load_state_dict(model_data)
Out[37]:
<All keys matched successfully>In [38]:
netG.eval()
Out[38]:
ResNetGenerator(
(model): Sequential(
(0): ReflectionPad2d((3, 3, 3, 3))
(1): Conv2d(3, 64, kernel_size=(7, 7), stride=(1, 1))
(2): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(5): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(6): ReLU(inplace=True)
(7): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
(8): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(9): ReLU(inplace=True)
(10): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(11): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(12): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(13): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(14): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(15): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(16): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(17): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(18): ResNetBlock(
(conv_block): Sequential(
(0): ReflectionPad2d((1, 1, 1, 1))
(1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(2): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(3): ReLU(inplace=True)
(4): ReflectionPad2d((1, 1, 1, 1))
(5): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1))
(6): InstanceNorm2d(256, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
)
)
(19): ConvTranspose2d(256, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
(20): InstanceNorm2d(128, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(21): ReLU(inplace=True)
(22): ConvTranspose2d(128, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), output_padding=(1, 1))
(23): InstanceNorm2d(64, eps=1e-05, momentum=0.1, affine=False, track_running_stats=False)
(24): ReLU(inplace=True)
(25): ReflectionPad2d((3, 3, 3, 3))
(26): Conv2d(64, 3, kernel_size=(7, 7), stride=(1, 1))
(27): Tanh()
)
)We're ready to load a random image of a horse and see what our generator produces
In [39]:
from PIL import Image
from torchvision import transforms
In [40]:
preprocess = transforms.Compose([transforms.Resize(256),
transforms.ToTensor()])
In [41]:
img = Image.open('horse.jpg')
img
Out[41]:
Let's pass it through preprocessing and turn it into a properly shaped variable.
In [42]:
img_t = preprocess(img)
batch_t = torch. unsqueeze(img_t, 0)
In [43]:
batch_out = netG(batch_t)
batch_out is now the output of the generator, which we can convert back to an image.
In [45]:
out_t = (batch_out.data.squeeze() + 1.0) / 2.0
out_img = transforms.ToPILImage()(out_t)
out_img.save('zebra.jpg')
out_img
Out[45]:
2.3 Pretrained network that describes scenes¶
In order to get firsthand experience with a model involving natural language,
we will use a pretrained image-captioning model, generously provided by Ruotian Luo.
When presented with a natural image, this kind of model generates a caption in English that describes the scene.
2.3.1 NeuralTalk2¶
pass
2.4 Torch Hub¶
pass
2.5 Conclusion¶
- A pretrained network is a model that has already been trained on a dataset. Such networks can typically produce useful results immediately after loading the network parameters.
- By knowing how to use a pretrained model, we can integrate a neural network into a project without having to design or train it
'deep learning > deep learning with pytorch_manning' 카테고리의 다른 글
| deep learning with pytorch_manning_ch.05 (1) | 2021.07.04 |
|---|---|
| deep learning with pytorch_manning_ch.04 (1) | 2021.07.01 |
| deep learning with pytorch_manning_ch.03 (1) | 2021.06.29 |
Comments