【深度学习感念】感受野计算-VGG16

(以VGG16为例计算深度卷积神经网络的感受野)

---

VGG16网络结构

---

感受野计算结果

---

感受野计算方法 

自定义脚本[1]

vgg_16 = [
	# 1
	[3, 1], [3, 1], [2, 2],
	# 2
	[3, 1],  [3, 1], [2, 2],
	# 3
	[3, 1], [3, 1], [3, 1], [2, 2],
	# 4
	[3, 1], [3, 1], [3, 1], [2, 2],
	# 5
	[3, 1], [3, 1], [3, 1], [2, 2],
	# fc6, fake convolutional layer   
	[7, 1]]
vgg16_layers = [
	"3x3 conv 64", "3x3 conv 64", "pool1",
	"3x3 conv 128", "3x3 conv 128", "pool2",
	"3x3 conv 256", "3x3 conv 256", "3x3 conv 256", "pool3",
	"3x3 conv 512", "3x3 conv 512", "3x3 conv 512", "pool4",
	"3x3 conv 512", "3x3 conv 512", "3x3 conv 512", "pool5",
	"7x7 fc"]
def cal_receptive_field(kspairs, layers=None):
	# K: composed kernel, also the receptive field 
	# S: composed stride
	K, S = 1, 1
	# H = 224
	if not layers:
		layers = range(len(kspairs))
	for layer, kspair in zip(layers, kspairs):
		k, s = kspair
		K = (k-1) * S + K
		S = S * s
		# H = H//s
		# iamge size {0}'.format(H)
		
		print('layer {:<15}: {} [{:3},{:2}]'.format(layer, kspair, K, S))
		
cal_receptive_field(vgg_16, vgg16_layers)

工具[2]

https://fomoro.com/research/article/receptive-field-calculator#3,1,1,SAME;3,1,1,SAME;2,2,1,SAME;3,1,1,SAME;3,1,1,SAME;2,2,1,SAME;3,1,1,SAME;3,1,1,SAME;3,1,1,SAME;2,2,1,SAME;3,1,1,SAME;3,1,1,SAME;3,1,1,SAME;2,2,1,SAME;3,1,1,SAME;3,1,1,SAME;3,1,1,SAME;2,2,1,SAME;7,1,1,VALID

参考资料

[1] http://zike.io/posts/calculate-receptive-field-for-vgg-16/

[2] https://fomoro.com/research/article/receptive-field-calculator#