-
Notifications
You must be signed in to change notification settings - Fork 0
/
localisation.py
199 lines (132 loc) · 6.1 KB
/
localisation.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
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
import matplotlib.pyplot as plt
import numpy as np
import torch
import torch.nn as nn
from torchvision import transforms
from torchvision.io.image import read_image
from torchvision.transforms.functional import normalize, resize
import torch.nn.functional as F
import torch.optim as optim
from PIL import Image
from torch.autograd import Variable
from torch import topk
import skimage.transform
from matplotlib.patches import Circle
from train import CNN
########################################################################################################
# Load Model:
########################################################################################################
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print('device : ', device)
image = Image.open('IMGs/TEST/P/84.jpg')
normalize = transforms.Normalize(
mean=[0.5, 0.5, 0.5],
std=[0.25, 0.25, 0.25]
)
preprocess = transforms.Compose([
transforms.Resize((224,224)),
transforms.ToTensor(),
normalize
])
display_transform = transforms.Compose([
transforms.Resize((224,224))])
tensor = preprocess(image)
prediction_var = Variable((tensor.unsqueeze(0)).cuda(), requires_grad=True)
# Using ResNET18 :
model = CNN().to(device)
PATH = 'checkpoints/FullNet.pth'
model.load_state_dict(torch.load(PATH))
model.cuda()
model.eval()
########################################################################################################
# Class Activation Map (CAM):
########################################################################################################
class SaveFeatures():
features = None
def __init__(self, m): self.hook = m.register_forward_hook(self.hook_fn)
def hook_fn(self, module, input, output): self.features = ((output.cpu()).data).numpy()
def remove(self): self.hook.remove()
final_layer = model._modules.get('layer4')
activated_features = SaveFeatures(final_layer)
prediction = model(prediction_var)
pred_probabilities = F.softmax(prediction).data.squeeze()
activated_features.remove()
topk(pred_probabilities, 1)
def getCAM(feature_conv, weight_fc, class_idx):
_, nc, h, w = feature_conv.shape
cam = weight_fc[class_idx].dot(feature_conv.reshape((nc, h*w)))
cam = cam.reshape(h, w)
cam = cam - np.min(cam)
cam_img = cam / np.max(cam)
return [cam_img]
weight_softmax_params = list(model._modules.get('fc').parameters())
weight_softmax = np.squeeze(weight_softmax_params[0].cpu().data.numpy())
weight_softmax_params
class_idx = topk(pred_probabilities, 1)[1].int()
overlay = getCAM(activated_features.features, weight_softmax, class_idx )
t = torch.tensor([ [[1.2,2.3],[3.4,4.5],[5.6,6.7]],[[1.2,2.3],[3.4,4.5],[5.6,6.7]] ])
u = torch.tensor( [[1.2, 2.3, 3.4, 4.5, 5.6, 6.7]] )
# a = t.unsqueeze(0)
# print(a)
# pred = Variable( a, requires_grad=True)
# print(pred)
print(activated_features)
topk( u, 1)
plt.figure(figsize=(20,20))
plt.subplot(141), plt.imshow(image.resize((224, 224)))
plt.subplot(142), plt.imshow(display_transform(image)), plt.imshow(skimage.transform.resize(overlay[0], tensor.shape[1:3]), alpha=0.7, cmap='plasma')
heat_map = skimage.transform.resize(overlay[0], tensor.shape[1:3])
plt.subplot(143), plt.imshow(heat_map)
# Position :
arr = np.array(heat_map)
max = np.where(arr == np.amax(arr))
x, y = max[0][0], max[1][0]
patch = [ Circle((y, x), radius=7, color='lime') ]
plt.subplot(144), plt.imshow(image.resize((224, 224))), plt.gca().add_patch(patch[0])
########################################################################################################
# Gradient basd method:
########################################################################################################
img = read_image('IMGs/TEST/P/1197.jpg')
tensor = normalize(resize(img, (224, 224)) / 255., [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]).unsqueeze(0).to(device)
tensor.requires_grad = True
out = model(tensor)
val = out.argmax(dim = 1)
gradients = torch.autograd.grad(outputs = out.squeeze(0)[val], inputs=tensor, retain_graph=True)[0]
heatmap = gradients[0][0] + gradients[0][1] + gradients[0][2]
print(heatmap.shape)
plt.figure(figsize = (20, 20))
plt.subplot(131) , plt.imshow( resize( img, (224, 224) ).permute(1,2,0) )
plt.subplot(132) , plt.imshow( skimage.transform.resize( heatmap.cpu() , (24, 24)) , alpha=0.9, cmap = 'RdYlBu')
arr = np.array(skimage.transform.resize(heatmap.cpu(), (224, 224)))
max = np.where(arr == np.amax(arr))
x, y = max[0][0], max[1][0]
patch = [ Circle((y, x), radius=7, color='yellow') ]
plt.subplot(133), plt.imshow( skimage.transform.resize( img.permute(1, 2, 0), (224, 224)) ), plt.gca().add_patch(patch[0])
plt.show()
"""
##################################################################################################################
# CAM: (another method)
##################################################################################################################
# img = read_image('TEST/P/5844.jpg')
# IMG to TENSOR :
tensor = normalize(resize(img, (224, 224)) / 255., [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]).unsqueeze(0).to(device)
out = model(tensor)
fm0 = model.maxpool( model.relu( model.bn1( model.conv1(tensor) ) ) )
fm4 = model.layer4( model.layer3( model.layer2( model.layer1( fm0 ) ) ) )
fmfc = fm4.reshape(512,7*7)
fn = model.fc(fmfc.permute(1,0)).permute(1,0).cpu().detach().numpy()
hm = fn.reshape(2,7,7)
plt.figure(figsize = (20, 20) )
plt.subplot(131), plt.imshow( skimage.transform.resize(img.permute(1,2,0), (224, 224)) )
plt.subplot(132), plt.imshow( skimage.transform.resize(hm[1], (224, 224)) , alpha=0.9, cmap = 'jet')
# plt.figure(figsize = (15, 15) )
# plt.subplot(121), plt.imshow( skimage.transform.resize( img.permute(1, 2, 0), (224, 224)) ),
# plt.imshow( skimage.transform.resize(hm[1], (224, 224)) , alpha=0.5, cmap='jet')
from skimage import io
from matplotlib.patches import Circle
arr = np.array(skimage.transform.resize(hm[1], (224, 224)))
max = np.where(arr == np.amax(arr))
x, y = max[0][0], max[1][0]
patch = [ Circle((y, x), radius=7, color='yellow') ]
plt.subplot(133), plt.imshow( skimage.transform.resize( img.permute(1, 2, 0), (224, 224)) ), plt.gca().add_patch(patch[0])
"""