SemSharing/second_client.py at main · chaof96/SemSharing · GitHub

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
# Import necessary libraries
import matplotlib.pyplot as plt
import scipy.io as io
from SuperGlue.models.superglue import SuperGlue
from SuperGlue.models.utils import make_matching_plot_fast
from api.ReconAnom import JSRNet_api
from api.SuperPoint import SuperPoint_api
import cv2
import os
import torch
import skimage
import numpy as np
import matplotlib.cm as cm

# Function to blend two images together with a specified blending factor
def blend(a, b, alpha=0.5):
    a = skimage.img_as_float(a)
    b = skimage.img_as_float(b)
    return a*alpha+(1-alpha)*b

class second_info():
    def __init__(self, weight='outdoor', match_threshold=0.2):
        # Set SuperGlue parameters
        superglue = weight
        sinkhorn_iterations = 30
        match_threshold = match_threshold
        self.match_threshold = match_threshold
        config = {
            'superglue': {
                'weights': superglue,
                'sinkhorn_iterations': sinkhorn_iterations,
                'match_threshold': match_threshold,
            }
        }
        # Initialize SuperGlue model with given configuration
        self.superglue = SuperGlue(config.get('superglue', {})).eval()
        self.superpoint = SuperPoint_api()

    # Perform matching of keypoints between two images
    def matching(self, sup_data_first, img2_path, size=(768, 768)):
        refer = cv2.imread(img2_path, 1)
        refer = cv2.resize(refer, size)
        sup_inp = cv2.cvtColor(refer, cv2.COLOR_BGR2GRAY) / 255.
        sup_res_refer = self.superpoint.run_SuperPoint(sup_inp, name='1')
        data = {**sup_data_first, **sup_res_refer}
        for k in data:
            if isinstance(data[k], (list, tuple)):
                data[k] = torch.stack(data[k])
        pred = {**data, **self.superglue(data)}
        return pred

    # Register keypoints between two images and return the transformed image
    def registration(self, pred, jsr_res=None, show_matching=True):
        # Converting tensors to numpy arrays for further processing
        pred = {k: v[0].cpu().numpy() for k, v in pred.items()}
        kpts0, kpts1 = pred['keypoints0'], pred['keypoints1']
        matches, conf = pred['matches0'], pred['matching_scores0']

        # Select the keypoints with valid matches
        valid = matches > -1
        mkpts0 = kpts0[valid]
        mkpts1 = kpts1[matches[valid]]
        mconf = conf[valid]

        # Filter the matched keypoints based on match threshold
        match_threshold = self.match_threshold
        mkpts0 = mkpts0[mconf > match_threshold]
        mkpts1 = mkpts1[mconf > match_threshold]

        # Create an array of matched points
        src_pts = np.float32(mkpts0).reshape(-1, 1, 2)
        dst_pts = np.float32(mkpts1).reshape(-1, 1, 2)

        # Compute the homography matrix if enough matched points are present
        M = None
        if len(src_pts) > 4:
            M, mask = cv2.findHomography(src_pts, dst_pts, cv2.LMEDS, 1)

        # Check if homography
        # matrix has been computed successfully
        if M is None:
            print('Registration failed')
            return None

        # Use homography to warp perspective
        align = cv2.warpPerspective(jsr_res, M, (jsr_res.shape[1], jsr_res.shape[0]), borderMode=cv2.BORDER_CONSTANT,
                                            borderValue=0)

        # Show the matches between the two images if specified
        if show_matching:
            query = pred['image0'].squeeze()
            refer = pred['image1'].squeeze()
            query = (cv2.cvtColor(query, cv2.COLOR_GRAY2RGB)*255).astype(np.uint8)
            refer = (cv2.cvtColor(refer, cv2.COLOR_GRAY2RGB)*255).astype(np.uint8)

            # Apply homography to the first image
            align_query = cv2.warpPerspective(query, M, (jsr_res.shape[1], jsr_res.shape[0]), borderMode=cv2.BORDER_CONSTANT,
                                            borderValue=0)

            color = cm.jet(mconf)

            # Create matching plot
            show1 = make_matching_plot_fast(
                query, refer, kpts0, kpts1, mkpts0, mkpts1, color,
                'SuperGlue')

            # Display matching plot
            plt.imshow(show1)
            plt.annotate('SuperGlue Results', (0, 0), color='red', fontsize=20)
            plt.show()

            # Save matching plot
            io.savemat('SuperGlueResults.mat', {'show1': show1})

            # Display original and transformed first image, and the reference image
            plt.imshow(query)
            plt.annotate('First Gray', (0, 0), color='red', fontsize=20)
            plt.show()

            show_align = blend(align_query, refer)
            qtensor = torch.from_numpy(align_query).permute(2, 0, 1)[None, ...] / 255.
            rtensor = torch.from_numpy(refer).permute(2, 0, 1)[None, ...] / 255.

            plt.imshow(show_align)
            plt.annotate('Matching Results', (0, 0), color='red', fontsize=20)
            plt.show()

            # Save matching results
            io.savemat('MatchingResults.mat', {'show_align': show_align})

            plt.imshow(align_query)
            plt.annotate('align_query', (0, 0), color='red', fontsize=20)
            plt.show()

            # Save warped first image
            io.savemat('align_query.mat', {'align_query': align_query})

            plt.imshow(refer)
            plt.annotate('refer', (0, 0), color='red', fontsize=20)
            plt.show()

            # Save reference image
            io.savemat('refer.mat', {'refer': refer})

        return align