-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathresnet_run.py
183 lines (147 loc) · 5.91 KB
/
resnet_run.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
import os
import torch
from torch import nn
import torch.utils.data as data
from torch.utils.data import DataLoader
import torchvision.transforms as transforms
from PIL import Image
import itertools
from Classifier import Classifier
from torchvision.models import resnet50
class DualBranchModel(nn.Module):
def __init__(self, backbone, is_train=True, input_size=1000, num_classes=701):
super(DualBranchModel, self).__init__()
self.device = device
self.backbone_ = backbone
self.classif = Classifier(num_classes=num_classes, input_size=input_size, is_train=is_train)
def forward(self, x1, x2):
x1 = self.backbone_(x1)
x2 = self.backbone_(x2)
x1 = self.classif(x1)
x2 = self.classif(x2)
return x1, x2
device = 'cuda' if torch.cuda.is_available() else 'cpu'
# device = 'cpu'
print(device, '\n')
backbone = resnet50().to(device)
model = DualBranchModel(backbone, is_train=False).to(device)
model.load_state_dict(torch.load('model_res.pt'))
model.eval()
# transform = transforms.Compose([
# transforms.Resize((256, 256)),
# transforms.ToTensor(),
# # transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
# ])
batch_size = 1
# img1 = Image.open(r'E:\Datasets\University-Release\test\gallery_drone\0001\image-06.jpeg')
# img1 = transform(img1).to(device)
# img2 = Image.open(r'E:\Datasets\University-Release\test\gallery_satellite\0004\0004.jpg')
# img2 = transform(img2).to(device)
# img3 = Image.open(r'E:\Datasets\University-Release\test\gallery_satellite\0005\0005.jpg')
# img3 = transform(img3).to(device)
# img4 = Image.open(r'E:\Datasets\University-Release\test\gallery_satellite\0001\0001.jpg')
# img4 = transform(img4).to(device)
# l = [(img1, img2, 0), (img1, img3, 0),
# (img1, img4, 1)]
# dataloader = DataLoader(l, batch_size=batch_size)
# input_tensor1 = torch.randn(3, 256, 256).to(device)
# input_tensor2 = torch.randn(3, 256, 256).to(device)
# input_tensor3 = torch.randn(3, 256, 256).to(device)
# input_tensor4 = torch.randn(3, 256, 256).to(device)
# l = [(input_tensor1, input_tensor2, 0), (input_tensor3, input_tensor4, 1),
# (input_tensor1, input_tensor4, 0), (input_tensor3, input_tensor2, 0)]
# dataloader = DataLoader(l, batch_size=batch_size)
class BuildingDataset(data.Dataset):
def __init__(self, root_dir, transform=None, mode='both'):
self.root_dir = root_dir
self.transform = transform
self.buildings = sorted(os.listdir(os.path.join(self.root_dir, 'test', 'gallery_drone')))
self.mode = mode
def __len__(self):
return len(self.buildings)
def __getitem__(self, index):
building_name = self.buildings[index]
drone_dir = os.path.join(self.root_dir, 'test', 'gallery_drone', building_name)
satellite_dir = os.path.join(self.root_dir, 'test', 'gallery_satellite', building_name)
satellite_path = os.path.join(satellite_dir, os.listdir(satellite_dir)[0])
satellite_img = Image.open(satellite_path).convert('RGB')
if self.mode == 'drone':
def drone_imgs():
for drone_img_name in os.listdir(drone_dir):
drone_img_path = os.path.join(drone_dir, drone_img_name)
drone_img = Image.open(drone_img_path).convert('RGB')
yield drone_img
if self.transform:
dataset = [self.transform(drone_img) for drone_img in drone_imgs()]
return dataset
if self.mode == 'satellite':
if self.transform:
satellite_img = self.transform(satellite_img)
return satellite_img
else:
def drone_imgs():
for drone_img_name in os.listdir(drone_dir):
drone_img_path = os.path.join(drone_dir, drone_img_name)
drone_img = Image.open(drone_img_path).convert('RGB')
yield drone_img
if self.transform:
satellite_img = self.transform(satellite_img)
dataset = [(satellite_img, self.transform(drone_img)) for drone_img in drone_imgs()]
return dataset
transform = transforms.Compose([
transforms.Resize((256, 256)),
transforms.ToTensor(),
# transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
dir = r'C:\Users\Insight\Documents\Обучение\Магистратура\2 семестр\Методы машинного обучения в робототехнике\Курсовая работа\University-Release'
dataset_satellite = BuildingDataset(root_dir=dir, transform=transform, mode='satellite')
dataset_drone = BuildingDataset(root_dir=dir, transform=transform, mode='drone')
dataset_satellite_loader = [dataset_satellite[i] for i in range(len(dataset_satellite))]
len_dataset = 35
result_dataset = []
# создаем спсиок для теста
for position in range(len_dataset):
vector = [0] * len_dataset
vector[position] = 1
for j in range(54):
dataset_drone_loader = [dataset_drone[position][j]] * len(dataset_drone)
result = [(x, y, z) for x, y, z in zip(dataset_satellite_loader[:len_dataset], dataset_drone_loader, vector)]
result_dataset.append(result)
print('Старт')
R1 = 0
R5 = 0
R10 = 0
count = 0
for j in range(len(result_dataset)):
dataloader = DataLoader(result_dataset[j], batch_size=batch_size)
distances = []
for i, (sat_img, uav_img, target) in enumerate(dataloader):
sat_img = sat_img.to(device)
uav_img = uav_img.to(device)
target = target.to(device)
sat, uav = model(sat_img, uav_img)
distance = torch.linalg.vector_norm(uav - sat, ord=2)
distances.append((distance.item(), i))
if target == 1:
true_index = i
sorted_distances = sorted(distances, key=lambda x: x[0])
for distance, idx in itertools.islice(sorted_distances, 5):
print(f"Distance: {distance}, i: {idx}")
print()
count += 1
if sorted_distances[0][1] == true_index:
R1 += 1
print('R1 +')
for d in range(5):
if sorted_distances[d][1] == true_index:
R5 += 1
print('R5 +')
break
for k in range(10):
if sorted_distances[k][1] == true_index:
R10 += 1
print('R10 +')
break
print(f'R1:{R1 / count}')
print(f'R5:{R5 / count}')
print(f'R10:{R10 / count}')