-
Notifications
You must be signed in to change notification settings - Fork 4
/
VisualBackProp.py
150 lines (118 loc) · 4.62 KB
/
VisualBackProp.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
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
import torch
import torch.nn as nn
from torch.nn.functional import conv_transpose2d
import torch.nn.functional as F
import torchvision
import torchvision.models as models
from torchvision.models import resnet18, resnet34, resnet101, resnet152
from torchvision import transforms
from torchvision.models.resnet import Bottleneck
from torch.utils.data import Dataset, DataLoader
from pathlib import Path
import pandas as pd
import numpy as np
from skimage import data, transform
from PIL import Image
import matplotlib.pyplot as plt
from os import walk
class ResnetVisualizer(nn.Module):
def __init__(self, resnet,weight_list):
super(ResnetVisualizer, self).__init__()
self.model = resnet
self.weight_list = weight_list
for name, child in self.model.named_children():
if 'layer' in name:
setattr(self, name, LayerVisualizer(name, child, weight_list))
# For Deconv
self.k7x7 = torch.ones((1,1,7,7))
self.k3x3 = torch.ones((1,1,3,3))
def forward(self, x):
input = x
x = self.model.conv1(x)
x = self.model.bn1(x)
x = self.model.relu(x)
act1 = x.mean(1, keepdim=True)
x = self.model.maxpool(x)
x, vis1 = self.layer1(x)
x, vis2 = self.layer2(x)
x, vis3 = self.layer3(x)
target_feature_map, vis4 = self.layer4(x)
## average pooling is better than max pooling
x = self.model.avgpool(target_feature_map)
vis = list(reversed([act1] + vis1 + vis2 + vis3 + vis4))
prod = vis[0]
for i in range(1, len(vis)):
act = vis[i]
if prod.shape != act.shape:
prod = conv_transpose2d(prod, self.k3x3, stride=2, padding=1, output_padding=1)
prod *= act
# Resize to input image
prod = conv_transpose2d(
prod, self.k7x7, stride=2, padding=3, output_padding=1)
return x.numpy().flatten(), prod, target_feature_map #* input.mean(1, keepdim=True)
class LayerVisualizer(nn.Module):
def __init__(self, name, layer, weight_list):
super(LayerVisualizer, self).__init__()
self.name = name
self.layer = layer
self.weight_list = weight_list
for name, child in self.layer.named_children():
if self.name == "layer4":
setattr(self, name, BlockVisualizer(name,child,True,weight_list))
else:
setattr(self, name, BlockVisualizer(name,child,False,weight_list))
def forward(self, x):
vis=[]
for name, child in self.layer.named_children():
block = getattr(self, name)
x, prod = block(x)
vis += prod
return x, vis
class BlockVisualizer(nn.Module):
def __init__(self,name, block, lastLayer,weight_list):
super(BlockVisualizer, self).__init__()
self.name = name
self.block = block
self.lastLayer = lastLayer
self.weight_list = weight_list
self.k3x3 = torch.ones((1,1,3,3))
self.k1x1 = torch.ones((1,1,1,1))
def forward(self, x):
vis = []
residual = x
out = self.block.conv1(x)
out = self.block.bn1(out)
out = self.block.relu(out)
vis += [out.mean(1,keepdim=True)]
out = self.block.conv2(out)
out = self.block.bn2(out)
if self.block.downsample is not None:
residual = self.block.downsample(x)
out += residual
out = self.block.relu(out)
if (self.lastLayer) == True and (self.name == '1'):
heatmap = (out.permute(0,2,3,1) * self.weight_list).permute(0,3,1,2)
heatmap = torch.mean(heatmap, 1, keepdim=True)
# heatmap_max = heatmap.max(axis = 0)[0]
# heatmap /= heatmap_max
# heatmap = heatmap.unsqueeze(0).unsqueeze(0)
vis += [heatmap]
else:
vis += [out.mean(1,keepdim=True)]
return out, [vis[-1]] #vis#[out.mean(1,keepdim=True)]
class FilenameDataset(Dataset):
def __init__(self, files, transform=None):
self.files = list(files)
self.transform = transform
def __len__(self):
return len(self.files)
def __getitem__(self, idx):
sample = Image.open(self.files[idx]).convert('RGB')
if self.transform:
return self.transform(sample)
transform_default = transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
return transform_default(sample)