- VGG常用来对图片进行特征提取，pytorch实现如下：
from torchvision import models
model = models.vgg16_bn(pretrained=True)
for param in model.parameters():
param.requires_grad = False
params_to_update = []
for name, param in self.model.named_parameters():
# 只训练需要的参数，比如最后的分类器
if param.requires_grad == True:
params_to_update.append(param)
print("\t", name)
得到的输出：
Params to learn:
classifier.weight
classifier.bias
- 对conv层输出的特征图片进行可视化
# visualization of feature map
model = models.vgg16_bn(pretrained=True)
print(model.children)
得到VGG的结构：
<bound method Module.children of VGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU(inplace=True)
(3): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(4): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(5): ReLU(inplace=True)
(6): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(7): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(9): ReLU(inplace=True)
(10): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(12): ReLU(inplace=True)
(13): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(14): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(16): ReLU(inplace=True)
(17): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(19): ReLU(inplace=True)
(20): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(21): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(22): ReLU(inplace=True)
(23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(24): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(26): ReLU(inplace=True)
(27): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(28): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(29): ReLU(inplace=True)
(30): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(31): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(32): ReLU(inplace=True)
(33): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(34): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(35): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(36): ReLU(inplace=True)
(37): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(38): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(39): ReLU(inplace=True)
(40): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(41): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(42): ReLU(inplace=True)
(43): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU(inplace=True)
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=4096, bias=True)
(4): ReLU(inplace=True)
(5): Dropout(p=0.5, inplace=False)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
)>
然后找到所有的卷积层
model_weights = [] # we will save the conv layer weights in this list
conv_layers = [] # we will save the conv layers in this list
# get all the model children as list
model_children = list(model.children())
# counter to keep count of the conv layers
counter = 0
# append all the conv layers and their respective weights to the list
for i in range(len(model_children)):
if type(model_children[i]) == nn.Conv2d:
counter += 1
model_weights.append(model_children[i].weight)
conv_layers.append(model_children[i])
elif type(model_children[i]) == nn.Sequential:
for j in range(len(model_children[i])):
child = model_children[i][j]
if type(child) == nn.Conv2d:
counter += 1
model_weights.append(child.weight)
conv_layers.append(child)
print(f"Total convolutional layers: {counter}")
得到结果 Total convolutional layers: 13
对特征图进行可视化
# visualize the first conv layer filters
plt.figure(figsize=(20, 17))
for i, filter in enumerate(model_weights[0]):
plt.subplot(int(len(model_weights[0]) / 8), 8,
i + 1) # (8, 8) because in conv0 we have 7x7 filters and total of 64 (see printed shapes)
plt.imshow(filter[0, :, :].detach(), cmap='gray')
plt.axis('off')
# plt.savefig('../outputs/filter.png')
plt.show()
得到vgg卷积层的filters的可视化
vgg的卷积核
接下来传入图片，看输入的图片经过卷积处理会变成什么样子。首先我们输入一只狗：
dogs = torch.utils.data.DataLoader(dogset, batch_size=1,
shuffle=False, num_workers=1)
img = next(iter(dogs))[0].to(device)
plt.imshow(np.transpose(img[0].cpu().numpy(), (1, 2, 0)))
plt.show()
得到了狗：
一只狗子
然后看看它通过卷积处理的样子：
# pass the image through all the layers
results = [conv_layers[0](img)]
for i in range(1, len(conv_layers)):
# pass the result from the last layer to the next layer
results.append(conv_layers[i](results[-1]))
# make a copy of the `results`
outputs = results
# visualize 64 features from each layer
# (although there are more feature maps in the upper layers)
for num_layer in range(len(outputs)):
fig = plt.figure(figsize=(30, 30))
layer_viz = outputs[num_layer][0, :, :, :]
layer_viz = layer_viz.data
print(layer_viz.size())
fig.suptitle(str(layer_viz.size()), fontsize=36)
for i, filter in enumerate(layer_viz):
if i == 16: # we will visualize only 4x4 blocks from each layer
break
plt.subplot(4, 4, i + 1)
plt.imshow(filter.cpu(), cmap='gray')
plt.axis("off")
print(f"Saving layer {num_layer} feature maps...")
plt.savefig(f"./output/features/dogr400_layer_{num_layer}.png")
plt.show()
plt.close()
展示一下其中的一张图，可以看到这是狗子的轮廓已经被提取出来了，但是还有一些细节在：
128x200x200
狗子浮雕画：
512x200x200
参考：
pytorch教程
Visualizing filters and feature maps in convolutional neural networks using pytorch
网友评论