-
Notifications
You must be signed in to change notification settings - Fork 0
/
c0_backbones.py
117 lines (103 loc) · 5.86 KB
/
c0_backbones.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
import math
import os
import cv2
import datetime
import random
import numpy as np
import pandas as pd
import tensorflow as tf
import keras
import keras.backend as K
import keras.layers as KL
import keras.engine as KE
import keras.models as KM
from keras.engine.saving import model_from_json,load_model
from a_config import Config
from b2_net_multi import BaseNetM
from image_set import ImageSet
from module import uu
from osio import mkdir_ifexist,to_excel_sheet
from postprocess import g_kern_rect,draw_text
# VGG 224x224 # kera-applications
## (/1) b1_c2 , (/2) b1_mp, b2_c2, (/4) b2_mp, b3_c3, (/8) b3_mp, b4_c3, (/16) b4_mp, b5_c3, (/32) b5_mp
## c1, p2, c2, p3, c3, p4, c4, p5, c5, p6 ## conv,pool
def keras_vgg_backbone(input_image,architecture,**kwargs):
from c1_vgg import NetU_Vgg
creator,convs=NetU_Vgg.config[architecture]
model=creator(input_tensor=input_image,include_top=False,pooling=None,**kwargs)
from module import dca,ca3,ca33
from module_complex import c3t2mp, c3t3mp
return [model.get_layer(name='block{}_pool'.format(bl+1)).output for bl,cv in enumerate(convs)]
# return [model.get_layer(name='block{}_pool'.format(bl+2)).output for bl,cv in enumerate(convs[1:])]+\
# [dca(model.get_layer(name='block{}_pool'.format(convs[-1]+2)).output,2,'dca%d'%(convs[-1]+2),6,512,'relu')]
# return [model.get_layer(name='block{}_conv{}'.format(bl+2,cv)).output for bl,cv in enumerate(convs[1:])]+\
# [ca33(model.get_layer(name='block{}_pool'.format(convs[-1]+2)).output,'f_ca%d'%(convs[-1]+1),6,512,'relu')]
def v16(input_image,**kwargs):
return keras_vgg_backbone(input_image,'vgg16',**kwargs) # 64x2 128x2 256x3 512x3 512x3
def v19(input_image,**kwargs):
return keras_vgg_backbone(input_image,'vgg19',**kwargs) # 64x2 128x2 256x4 512x4 512x4
# Resnet Graph 224x224 #
def res50(input_image,**kwargs):
# from model import resnet_graph
# return resnet_graph(input_image, "resnet50" ,stage5=True,train_bn=True)
# return resnet_graph(input_image, "resnet101" ,stage5=True,train_bn=True)
from c1_resnet import NetU_ResNet
creator,numbers=NetU_ResNet.config['resnet50']
model=creator(input_tensor=input_image,include_top=False,pooling=None,**kwargs)
return [model.get_layer(name='activation_%d'%n).output for n in numbers]
# DenseNet # kera-applications # @Fizyr # https://arxiv.org/pdf/1608.06993v3.pdf
def keras_densenet_backbone(input_image,architecture,**kwargs):
from c1_dense import NetU_Dense
creator,blocks=NetU_Dense.config[architecture]
model=creator(input_tensor=input_image,include_top=False,**kwargs)
print(model.summary())
return [model.get_layer(name='conv1/relu').output]+[model.get_layer(name='conv{}_block{}_concat'.format(idx+2,block_num)).output for idx,block_num in enumerate(blocks)]
# return [model.get_layer(name='conv1/relu').output]+[model.get_layer(name='pool{}_relu'.format(layer)).output for layer in range(2,5)]+[model.get_layer(name='relu').output]
def densenet121(input_image,**kwargs):
return keras_densenet_backbone(input_image,'densenet121',**kwargs)
def densenet169(input_image,**kwargs):
return keras_densenet_backbone(input_image,'densenet169',**kwargs)
def densenet201(input_image,**kwargs):
return keras_densenet_backbone(input_image,'densenet201',**kwargs)
# MobileNet #
def mobile(input_image,**kwargs):
from keras.applications import mobilenet
model=mobilenet.MobileNet(input_tensor=input_image,include_top=False,**kwargs)
return [model.get_layer(name=n).output for n in ['conv_pw_1_relu','conv_pw_3_relu','conv_pw_5_relu','conv_pw_11_relu','conv_pw_13_relu']]
## backbones not extracted correctly ##
# print(model.summary())
# for layer in model.layers:
# if layer.__class__.__name__=='Model':
# print("Model: ",layer.name)
# elif hasattr(layer,'strides') and layer.strides==(2,2): # catch down-sample layers
# print(layer.input,layer.output,sep='->',end='\n')
def mobile2(input_image,**kwargs):
from keras.applications import mobilenetv2
model=mobilenetv2.MobileNetV2(input_tensor=input_image,include_top=False,**kwargs)
return [KL.ZeroPadding2D(((1,0),(1,0)))(model.get_layer(name=n).output) for n in ['Conv1_pad','block_1_pad','block_3_pad','block_6_pad','block_13_pad']]
# Inception Xception # 299x299 hard to match dimension #
def incept3(input_image,**kwargs):
from keras.applications import inception_v3
model=inception_v3.InceptionV3(input_tensor=input_image,include_top=False,**kwargs)
return [KL.ZeroPadding2D(p)(model.get_layer(name=n).output) for p,n in [(((1,0),(1,0)),'activation_3'),(((1,0),(1,0)),'activation_5'),(((1,0),(1,0)),
'mixed2'),(((1,1),(1,1)),'mixed7'),(((0,1),(0,1)),'mixed10')]]
def incepres2(input_image,**kwargs):
from keras.applications import inception_resnet_v2
model=inception_resnet_v2.InceptionResNetV2(input_tensor=input_image,include_top=False,**kwargs)
print(model.summary())
return [KL.ZeroPadding2D(p)(model.get_layer(name=n).output) for p,n in [(((1,1),(1,1)),'conv2d_1'),(((1,2),(1,2)),'max_pooling2d_1'),(((1,2),(2,1)),
'max_pooling2d_2'),(((1,1),(1,1)),'max_pooling2d_3'),(((1,1),(1,1)),'max_pooling2d_4')]]
def xcept(input_image,**kwargs):
from keras.applications import xception
model=xception.Xception(input_tensor=input_image,include_top=False,**kwargs)
print(model.summary())
return model
# NASNet #
def nasmobile(input_image,**kwargs):
from keras.applications import nasnet
model=nasnet.NASNetMobile(input_tensor=input_image,include_top=False,**kwargs)
return model
def naslarge(input_image,**kwargs):
from keras.applications import nasnet
model=nasnet.NASNetLarge(input_tensor=input_image,include_top=False,**kwargs)
return model