Medical-image-segment

Segment network architecture

refernence paper:

《The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation》

Simon Jégou, Michal Drozdzal, David Vazquez, Adriana Romero, Yoshua Bengio

input size:512x512

use 7x7 conv and 2x2 stride down sample to 256x256

use 2x2 maxpooling down sample to 128x128

output size:128x128，interpolation—>512x512

Dense block groth rate:24

Layer	feature map	filter	Conv	DB size
Input	512x512	1	NULL
Conv0	256x256	48	7x7
Maxpooling	128x128	48	2x2
DB1	128x128	120	3x3	3
TD1	64x64	120	1x1
DB2	64x64	240	3x3	5
TD2	32x32	240	1x1
DB3	32x32	456	3x3	7
TD3	16x16	456	1x1
Center	16x16	672	3x3	9
TU5	32x32	456	2x2
C5(TU5,DB3)	32x32	912	NULL
DB5	32x32	456	3x3	7
TU6	64x64	240	2x2
C6(TU6,DB2)	64x64	480	NULL
DB6	64x64	240	3x3	5
TU7	128x128	120	2x2
C7(TU7,DB1)	128x128	240	NULL
DB7	128x128	120	3x3	3
Output	128x128	1	1x1

layer:51

total size of network:26Mb

Name		Name	Last commit message	Last commit date
Latest commit History 38 Commits
.idea		.idea
__pycache__		__pycache__
data		data
logs		logs
reference		reference
result		result
static		static
temp		temp
weights		weights
.gitattributes		.gitattributes
AI.py		AI.py
Flask_serving.py		Flask_serving.py
LICENSE		LICENSE
NII_to_jpeg.py		NII_to_jpeg.py
README.md		README.md
inference_test.py		inference_test.py
loss.py		loss.py
mask.py		mask.py
read_data.py		read_data.py
read_test.py		read_test.py
test3.py		test3.py
tiramasu.py		tiramasu.py
tiramasu56.py		tiramasu56.py
tiramasu56_nodropout.py		tiramasu56_nodropout.py
tiramasu_test.py		tiramasu_test.py
train.py		train.py
unet_bn.py		unet_bn.py
unet_dense.py		unet_dense.py
unet_dense2.py		unet_dense2.py
unet_res.py		unet_res.py
unet_vgg.py		unet_vgg.py
visualize.py		visualize.py
www.py		www.py