GoogLeNet incarnation of the Inception architecture (Receptive Field Calculation)

Where K = Kernel size, P= Padding, S= Stride

η_in = Number Of Input Feature, η_out = Number Of Output Feature

R_in = Input receptive Field, R_out = Output Receptive field

J_in = Input Jump, J_out = Output Jump(increase in jump)

Layers	K	P	S	η_in	η_out= (η_in+2P-K)/S +1	J_in	J_out=J_in*S	R_in	R_out= R_in + (K-1)*J_in
convolution	7	2	2	224	112	1	2	1	1+(7-1)*1=7
max_pool	3	0	2	112	56	2	4	7	7+(3-1)*2=11
convolution	3	1	1	56	56	4	4	11	11+(3-1)*4=19
max_pool	3	0	2	56	28	4	8	19	19+(3-1)*4=27
inception (3a)	5	2	1	28	28	8	8	27	27+(5-1)*8=59
inception (3b)	5	2	1	28	28	8	8	59	59+(5-1)*8=91
max_pool	3	0	2	28	14	8	16	91	91+(3-1)*8=123
inception (4a)	5	2	1	14	14	16	16	123	123+(5-1)*16=187
inception (4b)	5	2	1	14	14	16	16	187	187+(5-1)*16=251
inception (4c)	5	2	1	14	14	16	16	251	251+(5-1)*16=315
inception (4d)	5	2	1	14	14	16	16	315	315+(5-1)*16=379
inception (4e)	5	2	1	14	14	16	16	379	379+(5-1)*16=443
max_pool	3	0	2	14	7	16	32	443	443+(3-1)*16=475
inception (5a)	5	2	1	7	7	32	32	475	475+(5-1)*32=603
inception (5b)	5	2	1	7	7	32	32	603	603+(5-1)*32=731
avg pool	7	1	1	1	1	32	32	731	731+(7-1)*32=923
dropout (40%)	-	-	-	-	-	-	-	-	-
linear	-	-	-	-	-	-	-	-	-
softmax	-	-	-	-	-	-	-	-	-

Calculate Receptive Field for VGG16.

Where K = Kernel size, P= Padding, S= Stride

η_in = Number Of Input Feature, η_out = Number Of Output Feature

R_in = Input receptive Field, R_out = Output Receptive field

J_in = Input Jump, J_out = Output Jump

Layers	K	P	S	η_in	η_out= (η_in+2P-K)/S +1	J_in	J_out=J_in*S	R_in	R_out= R_in + (K-1)*J_in
conv	3	1	1	224	(224+2-3)/1+1=224	1	1	1	1+(3-1)*1=3
conv	3	1	1	224	(224+2-3)/1+1=224	1	1	3	3+(3-1)*1=5
MP_1	2	0	2	224	(224+0-2)/2+1=112	1	2	5	5+(2-1)*2=6
conv	3	1	1	112	(112+2-3)/1+1=112	2	2	6	6+(3-1)*2=10
conv	3	1	1	112	(112+2-3)/1+1=112	2	2	10	10+(3-1)*2=14
MP_2	2	0	2	112	(112+0-2)/2+1=56	2	4	14	14+(2-1)*2=16
conv	3	1	1	56	(56+2-3)/1+1=56	4	4	16	16+(3-1)*4=24
conv	3	1	1	56	(56+2-3)/1+1=56	4	4	24	24+(3-1)*4=32
conv	3	1	1	56	(56+2-3)/1+1=56	4	4	32	32+(3-1)*4=40
MP_3	2	0	2	56	(56+0-2)/2+1=28	4	8	40	40+(2-1)*8=44
conv	3	1	1	28	(28+2-3)/1+1=28	8	8	44	44+(3-1)*8=60
conv	3	1	1	28	(28+2-3)/1+1=28	8	8	60	60+(3-1)*8=76
conv	3	1	1	28	(28+2-3)/1+1=28	8	8	76	76+(3-1)*8=92
MP_4	2	0	2	28	(28+0-2)/2+1=14	8	16	92	92+(2-1)*8=100
conv	3	1	1	14	(14+2-3)/1+1=14	16	16	100	100+(3-1)*16=132
conv	3	1	1	14	(14+2-3)/1+1=14	16	16	132	132+(3-1)*16=164
conv	3	1	1	14	(14+2-3)/1+1=14	16	16	164	164+(3-1)*16=196
MP_5	2	0	2	14	(14+0-2)/2+1=7	16	32	196	196+(2-1)*16=212
FC	7	0	1	7	(7+0-7)/1+1=1	32	32	212	212+(7-1)*32=404

Name		Name	Last commit message	Last commit date
Latest commit History 37 Commits
AWS_Deployment		AWS_Deployment
Facial_Landmark_Detection		Facial_Landmark_Detection
GAN		GAN
images		images
img		img
1st_DNN.ipynb		1st_DNN.ipynb
2nd_DNN_1.ipynb		2nd_DNN_1.ipynb
2nd_DNN_2.ipynb		2nd_DNN_2.ipynb
2nd_DNN_3.ipynb		2nd_DNN_3.ipynb
2nd_DNN_4.ipynb		2nd_DNN_4.ipynb
4.md		4.md
7.md		7.md
9A.ipynb		9A.ipynb
9B.ipynb		9B.ipynb
Assign_7(ENAS).ipynb		Assign_7(ENAS).ipynb
C(capture)_Profile+suffleNet.ipynb		C(capture)_Profile+suffleNet.ipynb
Colab_DN.ipynb		Colab_DN.ipynb
DN_with_aug.ipynb		DN_with_aug.ipynb
EVA_Assign_6A.ipynb		EVA_Assign_6A.ipynb
EVA_Assign_6A.py		EVA_Assign_6A.py
EVA_Assign_6B.ipynb		EVA_Assign_6B.ipynb
EVA_S2.ipynb		EVA_S2.ipynb
EVA_assign_5.ipynb		EVA_assign_5.ipynb
EVA_s1.ipynb		EVA_s1.ipynb
EVA_s1.md		EVA_s1.md
Readme.md		Readme.md
ResNet18_with_OneCycleLR.ipynb		ResNet18_with_OneCycleLR.ipynb
TFRecord_DNST_CIFAR10_AUG.ipynb		TFRecord_DNST_CIFAR10_AUG.ipynb
Unet.ipynb		Unet.ipynb
VGG16 imagenet weights visualization for each class activation.ipynb		VGG16 imagenet weights visualization for each class activation.ipynb
command's		command's
course-note.pdf		course-note.pdf
k-means_annotation_box.ipynb		k-means_annotation_box.ipynb
vgg_annotation_tool.html		vgg_annotation_tool.html