Computational Analysis and Deep Learning for Medical Care. Группа авторов

Чтение книги онлайн.

СКАЧАТЬ can be improved by increasing depth. Once the network gets converged, its accuracy saturates. Further, if we add more layers, then the performance gets degraded rapidly, which, in turn, results in higher training error. To solve the problem of the vanishing/exploding gradient, ResNet with a residual learning framework [6] was proposed by allowing new layers to fit a residual mapping. When a model is converged than to fit the mapping, it is easy to push the residual to zero. The principle of ResNet is residual learning and identity mapping and skip connections. The idea behind the residual learning is that it feeds the input image to the next convolutional layer and adds them together and performs non-linear activation (ReLU) and pooling.

Table 1.6 Various parameters of GoogleNet.

Layer name	Input size	Filter size	Window size	# Filters	Stride	Depth	# 1 × 1	# 3 × 3 reduce	# 3 × 3	# 5 × 5 reduce	# 5 × 5	Pool proj	Padding	Output size	Params	Ops
Convolution	224 × 224	7 × 7	-	64	2	1							2	112 × 112 × 64	2.7M	34M
Max pool	112 × 112	-	3 × 3	-	2	0							0	56 × 56 × 64
Convolution	56 × 56	3 × 3	-	192	1	2		64	192				1	56 × 56 × 192	112K	360M
Max pool	56 × 56	-	3 × 3	192	2	0							0	28 × 28 × 192
Inception (3a)	28 × 28	-	-	-	-	2	64	96	128	16	32	32	-	28 × 28 × 256	159K	128M
Inception (3b)	28 × 28	-	-	-	-	2	128	128	192	32	96	64	-	28 × 28 × 480	380K	304M
Max pool	28 × 28	-	3 × 3	480	2	0							0	14 × 14 × 480
Inception (4a)	14 × 14	-	-	-	-	2	192	96	208	16	48	64	-	14 × 14 × 512	364K	73M
Inception (4b)	14 × 14	-	-	-	-	2	160	112	224	24 СКАЧАТЬ В начало < 13 14 15 16 17 18 19 20 21 22 > В конец