/
main_demo.py
186 lines (156 loc) · 7.18 KB
/
main_demo.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
# !/usr/bin/env python
# -*- coding: utf-8 -*-
"""
...
"""
__author__ = "..."
__email__ = "..."
__license__ = "..."
__version__ = "1.0"
# External modules
from torchvision import transforms
import numpy as np
import cv2 as cv
import argparse
import torch
from PIL import Image
# Internal modules
from model.corrnet import CorrNet
from model.guided_backprop import GuidedBackpropReLUModel
from model import dip
from model import guided_grad_cam as guided_gc
def main(path_2_reference, path_2_target, path_2_corrnet, path_2_lcorrnet, device_id, batch_size, top_k):
# Variables
list_images = []
list_keypoints = []
list_keypoints_cv = []
list_descriptions = []
similarity_score = 0
mask = None
# Device
if device_id < 0:
print('Demo will run on CPU.')
device = torch.device('cpu')
else:
print('Demo will run on CUDA: {}'.format(device_id))
device = torch.device('cuda:{}'.format(device_id))
# Load CorNet
corrnet = CorrNet(feature_dim=512)
corrnet.load_state_dict(torch.load(path_2_corrnet, map_location=device))
corrnet = corrnet.to(device)
corrnet.eval()
# Load CorNet
l_corrnet = CorrNet(feature_dim=512)
l_corrnet.load_state_dict(torch.load(path_2_lcorrnet, map_location=device))
l_corrnet = l_corrnet.to(device)
l_corrnet.eval()
# Initialize guided backpropagation
guided_bp = GuidedBackpropReLUModel(corrnet, device)
# Transforms for evaluation
transform = transforms.Compose([transforms.Resize((240, 320)), transforms.ToTensor()])
# Load reference image
ref_img_np = dip.load_image(path_2_reference)
ref_img_pil = Image.fromarray(np.uint8(ref_img_np))
ref_img_tensor = transform(ref_img_pil).unsqueeze(0)
list_images.append(ref_img_np)
# Load target image
if path_2_target is None:
tar_img_tensor = None
else:
tar_img_np = dip.load_image(path_2_target)
tar_img_pil = Image.fromarray(np.uint8(tar_img_np))
tar_img_tensor = transform(tar_img_pil).unsqueeze(0)
list_images.append(tar_img_np)
# Run the CorrNet framework
if tar_img_tensor is None:
# Prepare input images
x = ref_img_tensor
x = x.to(device)
x.requires_grad = True
# Detect keypoints
ref_keypoints = dip.detect_keypoints(x, guided_bp, None, top_k=top_k)
list_keypoints.append(ref_keypoints)
else:
# Prepare input images
x = torch.cat([ref_img_tensor, tar_img_tensor])
x = x.to(device)
x.requires_grad = True
# Send input images to CorrNet
h, _ = corrnet(x)
# Compute highest-activated output neuron
highest_activated_neuron_id = torch.argmax(torch.mul(h[0], h[1]))
# Compute similarity score
similarity_score = np.clip(torch.mm(h, h.t().contiguous()).detach().cpu().numpy()[0, 1], 0., 1.)
# Detect keypoints
tar_img_tensor = tar_img_tensor.to(device)
tar_img_tensor.requires_grad = True
ref_img_tensor = ref_img_tensor.to(device)
ref_img_tensor.requires_grad = True
list_keypoints = guided_gc.detect_keypoints(ref_img_tensor, tar_img_tensor, corrnet, guided_bp, highest_activated_neuron_id, top_k=top_k)
# Extract descriptions
for i in range(2):
descriptions = dip.extract_descriptions(x[i].unsqueeze(0), list_keypoints[i], l_corrnet, device, batch_size)
list_descriptions.append(descriptions)
# Convert to OpenCV format
for keypoints in list_keypoints:
k_cv = []
for x, y, _ in keypoints:
k_cv.append(cv.KeyPoint(float(y), float(x), None))
list_keypoints_cv.append(k_cv)
# Show image
if len(list_images) > 1:
# Find correspondences
correspondences = dip.find_correspondences(list_descriptions)
# Compute homography
homo_matrix, homo_mask = dip.compute_homography(list_keypoints_cv, correspondences)
# Draw correspondences
if homo_matrix is None:
dip.draw_keypoints(list_images[0], list_keypoints[0])
dip.draw_keypoints(list_images[1], list_keypoints[1])
image_2_show = cv.cvtColor(np.hstack([list_images[0], list_images[1]]), cv.COLOR_RGB2BGR)
else:
h, w, _ = ref_img_np.shape
mask = homo_mask.ravel().tolist()
image_2_show_ref = cv.cvtColor(list_images[0], cv.COLOR_RGB2BGR)
image_2_show_tar = cv.cvtColor(list_images[1], cv.COLOR_RGB2BGR)
draw_params = dict(matchColor=(0, 255, 0), matchesMask=mask, singlePointColor=(0, 255, 0), flags=2)
image_2_show = cv.drawMatches(image_2_show_ref, list_keypoints_cv[0], image_2_show_tar, list_keypoints_cv[1], correspondences, None, **draw_params)
# Add text
h, w, c = image_2_show.shape
text_padding = 40
image_2_show_w_text = np.zeros((h + text_padding, w, c), dtype=np.uint8)
image_2_show_w_text[text_padding:, :, :] = image_2_show[:]
image_2_show = image_2_show_w_text
text = 'Sim.: {:.1f}% - # Kps. (r): {} - # Kps (t): {} - # Corrs.: {}'.format(similarity_score*100,
list_keypoints[0].shape[0],
list_keypoints[1].shape[0],
np.sum(mask))
cv.putText(image_2_show, text, (5, 25), cv.FONT_HERSHEY_SIMPLEX, .5, (255, 255, 255), 1, cv.LINE_AA)
else:
image_2_show = cv.cvtColor(list_images[0], cv.COLOR_RGB2BGR)
dip.draw_keypoints(image_2_show, list_keypoints[0], (0, 255, 0))
# Add text
h, w, c = image_2_show.shape
text_padding = 40
image_2_show_w_text = np.zeros((h + text_padding, w, c), dtype=np.uint8)
image_2_show_w_text[text_padding:, :, :] = image_2_show[:]
image_2_show = image_2_show_w_text
text = '# Kps. (r): {}'.format(list_keypoints[0].shape[0])
cv.putText(image_2_show, text, (5, 25), cv.FONT_HERSHEY_SIMPLEX, .5, (255, 255, 255), 1, cv.LINE_AA)
cv.imshow('Output', image_2_show)
cv.waitKey(0)
cv.destroyAllWindows()
if __name__ == '__main__':
# Parse arguments
parser = argparse.ArgumentParser(description='Demonstration of keypoint detection and description extraction with the CorrNet framework.')
parser.add_argument('-r', type=str, help='Reference image.', required=True)
parser.add_argument('-t', type=str, help='Target image.')
parser.add_argument('-c', type=str, help='Full path to CorrNet.', default='./corrnet.pth')
parser.add_argument('-l', type=str, help='Full path to l-CorrNet.', default='./l-corrnet.pth')
parser.add_argument('-d', type=int, help='CUDA id. The default running device is CPU.', default=-1)
parser.add_argument('-b', type=int, help='Batch size of l-CorrNet.', default=32)
parser.add_argument('-k', type=int, help='Top k keypoints.', default=1000)
args = parser.parse_args()
# Run demo
main(args.r, args.t, args.c, args.l, args.d, args.b, args.k)
exit(0)