File size: 10,621 Bytes
63f3cf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
# -*- coding: UTF-8 -*-
'''=================================================
@Project -> File   pram -> gm
@IDE    PyCharm
@Author fx221@cam.ac.uk
@Date   07/02/2024 10:47
=================================================='''
import torch
import torch.nn as nn
import torch.nn.functional as F
from nets.layers import KeypointEncoder, AttentionalPropagation
from nets.utils import normalize_keypoints, arange_like

eps = 1e-8


def dual_softmax(M, dustbin):
    M = torch.cat([M, dustbin.expand([M.shape[0], M.shape[1], 1])], dim=-1)
    M = torch.cat([M, dustbin.expand([M.shape[0], 1, M.shape[2]])], dim=-2)
    score = torch.log_softmax(M, dim=-1) + torch.log_softmax(M, dim=1)
    return torch.exp(score)


def sinkhorn(M, r, c, iteration):
    p = torch.softmax(M, dim=-1)
    u = torch.ones_like(r)
    v = torch.ones_like(c)
    for _ in range(iteration):
        u = r / ((p * v.unsqueeze(-2)).sum(-1) + eps)
        v = c / ((p * u.unsqueeze(-1)).sum(-2) + eps)
    p = p * u.unsqueeze(-1) * v.unsqueeze(-2)
    return p


def sink_algorithm(M, dustbin, iteration):
    M = torch.cat([M, dustbin.expand([M.shape[0], M.shape[1], 1])], dim=-1)
    M = torch.cat([M, dustbin.expand([M.shape[0], 1, M.shape[2]])], dim=-2)
    r = torch.ones([M.shape[0], M.shape[1] - 1], device='cuda')
    r = torch.cat([r, torch.ones([M.shape[0], 1], device='cuda') * M.shape[1]], dim=-1)
    c = torch.ones([M.shape[0], M.shape[2] - 1], device='cuda')
    c = torch.cat([c, torch.ones([M.shape[0], 1], device='cuda') * M.shape[2]], dim=-1)
    p = sinkhorn(M, r, c, iteration)
    return p


class AttentionalGNN(nn.Module):
    def __init__(self, feature_dim: int, layer_names: list, hidden_dim: int = 256, ac_fn: str = 'relu',
                 norm_fn: str = 'bn'):
        super().__init__()
        self.layers = nn.ModuleList([
            AttentionalPropagation(feature_dim=feature_dim, num_heads=4, hidden_dim=hidden_dim, ac_fn=ac_fn,
                                   norm_fn=norm_fn)
            for _ in range(len(layer_names))])
        self.names = layer_names

    def forward(self, desc0, desc1):
        # desc0s = []
        # desc1s = []

        for i, (layer, name) in enumerate(zip(self.layers, self.names)):
            if name == 'cross':
                src0, src1 = desc1, desc0
            else:
                src0, src1 = desc0, desc1
            delta0 = layer(desc0, src0)
            # prob0 = layer.attn.prob
            delta1 = layer(desc1, src1)
            # prob1 = layer.attn.prob
            desc0, desc1 = (desc0 + delta0), (desc1 + delta1)

            # if name == 'cross':
            #     desc0s.append(desc0)
            #     desc1s.append(desc1)
        return [desc0], [desc1]

    def predict(self, desc0, desc1, n_it=-1):
        for i, (layer, name) in enumerate(zip(self.layers, self.names)):
            if name == 'cross':
                src0, src1 = desc1, desc0
            else:
                src0, src1 = desc0, desc1
            delta0 = layer(desc0, src0)
            # prob0 = layer.attn.prob
            delta1 = layer(desc1, src1)
            # prob1 = layer.attn.prob
            desc0, desc1 = (desc0 + delta0), (desc1 + delta1)

            if name == 'cross' and i == n_it:
                break
        return [desc0], [desc1]


class GM(nn.Module):
    default_config = {
        'descriptor_dim': 128,
        'hidden_dim': 256,
        'keypoint_encoder': [32, 64, 128, 256],
        'GNN_layers': ['self', 'cross'] * 9,  # [self, cross, self, cross, ...] 9 in total
        'sinkhorn_iterations': 20,
        'match_threshold': 0.2,
        'with_pose': False,
        'n_layers': 9,
        'n_min_tokens': 256,
        'with_sinkhorn': True,

        'ac_fn': 'relu',
        'norm_fn': 'bn',
        'weight_path': None,
    }

    required_inputs = [
        'image0', 'keypoints0', 'scores0', 'descriptors0',
        'image1', 'keypoints1', 'scores1', 'descriptors1',
    ]

    def __init__(self, config):
        super().__init__()
        self.config = {**self.default_config, **config}
        print('gm: ', self.config)

        self.n_layers = self.config['n_layers']

        self.with_sinkhorn = self.config['with_sinkhorn']
        self.match_threshold = self.config['match_threshold']

        self.sinkhorn_iterations = self.config['sinkhorn_iterations']
        self.kenc = KeypointEncoder(
            self.config['descriptor_dim'] if self.config['descriptor_dim'] > 0 else 128,
            self.config['keypoint_encoder'],
            ac_fn=self.config['ac_fn'],
            norm_fn=self.config['norm_fn'])
        self.gnn = AttentionalGNN(
            feature_dim=self.config['descriptor_dim'] if self.config['descriptor_dim'] > 0 else 128,
            hidden_dim=self.config['hidden_dim'],
            layer_names=self.config['GNN_layers'],
            ac_fn=self.config['ac_fn'],
            norm_fn=self.config['norm_fn'],
        )

        self.final_proj = nn.ModuleList([nn.Conv1d(
            self.config['descriptor_dim'] if self.config['descriptor_dim'] > 0 else 128,
            self.config['descriptor_dim'] if self.config['descriptor_dim'] > 0 else 128,
            kernel_size=1, bias=True) for _ in range(self.n_layers)])

        bin_score = torch.nn.Parameter(torch.tensor(1.))
        self.register_parameter('bin_score', bin_score)

        self.match_net = None  # GraphLoss(config=self.config)

        self.self_prob0 = None
        self.self_prob1 = None
        self.cross_prob0 = None
        self.cross_prob1 = None

        self.desc_compressor = None

    def forward_train(self, data):
        pass

    def produce_matches(self, data, p=0.2, n_it=-1, **kwargs):
        kpts0, kpts1 = data['keypoints0'], data['keypoints1']
        scores0, scores1 = data['scores0'], data['scores1']
        if kpts0.shape[1] == 0 or kpts1.shape[1] == 0:  # no keypoints
            shape0, shape1 = kpts0.shape[:-1], kpts1.shape[:-1]
            return {
                'matches0': kpts0.new_full(shape0, -1, dtype=torch.int)[0],
                'matches1': kpts1.new_full(shape1, -1, dtype=torch.int)[0],
                'matching_scores0': kpts0.new_zeros(shape0)[0],
                'matching_scores1': kpts1.new_zeros(shape1)[0],
                'skip_train': True
            }

        if 'norm_keypoints0' in data.keys() and 'norm_keypoints1' in data.keys():
            norm_kpts0 = data['norm_keypoints0']
            norm_kpts1 = data['norm_keypoints1']
        elif 'image0' in data.keys() and 'image1' in data.keys():
            norm_kpts0 = normalize_keypoints(kpts0, data['image0'].shape)
            norm_kpts1 = normalize_keypoints(kpts1, data['image1'].shape)
        elif 'image_shape0' in data.keys() and 'image_shape1' in data.keys():
            norm_kpts0 = normalize_keypoints(kpts0, data['image_shape0'])
            norm_kpts1 = normalize_keypoints(kpts1, data['image_shape1'])
        else:
            raise ValueError('Require image shape for keypoint coordinate normalization')

        # Keypoint MLP encoder.
        enc0, enc1 = self.encode_keypoint(norm_kpts0=norm_kpts0, norm_kpts1=norm_kpts1, scores0=scores0,
                                          scores1=scores1)

        if self.config['descriptor_dim'] > 0:
            desc0, desc1 = data['descriptors0'], data['descriptors1']
            desc0 = desc0.transpose(0, 2, 1)  # [B, N, D ] -> [B, D, N]
            desc1 = desc1.transpose(0, 2, 1)  # [B, N, D ] -> [B, D, N]
            with torch.no_grad():
                if desc0.shape[1] != self.config['descriptor_dim']:
                    desc0 = self.desc_compressor(desc0)
                if desc1.shape[1] != self.config['descriptor_dim']:
                    desc1 = self.desc_compressor(desc1)
            desc0 = desc0 + enc0
            desc1 = desc1 + enc1
        else:
            desc0 = enc0
            desc1 = enc1

        desc0s, desc1s = self.gnn.predict(desc0, desc1, n_it=n_it)

        mdescs0 = self.final_proj[n_it](desc0s[-1])
        mdescs1 = self.final_proj[n_it](desc1s[-1])
        dist = torch.einsum('bdn,bdm->bnm', mdescs0, mdescs1)
        if self.config['descriptor_dim'] > 0:
            dist = dist / self.config['descriptor_dim'] ** .5
        else:
            dist = dist / 128 ** .5
        score = self.compute_score(dist=dist, dustbin=self.bin_score, iteration=self.sinkhorn_iterations)

        indices0, indices1, mscores0, mscores1 = self.compute_matches(scores=score, p=p)

        output = {
            'matches0': indices0,  # use -1 for invalid match
            'matches1': indices1,  # use -1 for invalid match
            'matching_scores0': mscores0,
            'matching_scores1': mscores1,
        }

        return output

    def forward(self, data, mode=0):
        if not self.training:
            return self.produce_matches(data=data, n_it=-1)
        return self.forward_train(data=data)

    def encode_keypoint(self, norm_kpts0, norm_kpts1, scores0, scores1):
        return self.kenc(norm_kpts0, scores0), self.kenc(norm_kpts1, scores1)

    def compute_distance(self, desc0, desc1, layer_id=-1):
        mdesc0 = self.final_proj[layer_id](desc0)
        mdesc1 = self.final_proj[layer_id](desc1)
        dist = torch.einsum('bdn,bdm->bnm', mdesc0, mdesc1)
        dist = dist / self.config['descriptor_dim'] ** .5
        return dist

    def compute_score(self, dist, dustbin, iteration):
        if self.with_sinkhorn:
            score = sink_algorithm(M=dist, dustbin=dustbin,
                                   iteration=iteration)  # [nI * nB, N, M]
        else:
            score = dual_softmax(M=dist, dustbin=dustbin)
        return score

    def compute_matches(self, scores, p=0.2):
        max0, max1 = scores[:, :-1, :-1].max(2), scores[:, :-1, :-1].max(1)
        indices0, indices1 = max0.indices, max1.indices
        mutual0 = arange_like(indices0, 1)[None] == indices1.gather(1, indices0)
        mutual1 = arange_like(indices1, 1)[None] == indices0.gather(1, indices1)
        zero = scores.new_tensor(0)
        # mscores0 = torch.where(mutual0, max0.values.exp(), zero)
        mscores0 = torch.where(mutual0, max0.values, zero)
        mscores1 = torch.where(mutual1, mscores0.gather(1, indices1), zero)
        # valid0 = mutual0 & (mscores0 > self.config['match_threshold'])
        valid0 = mutual0 & (mscores0 > p)
        valid1 = mutual1 & valid0.gather(1, indices1)
        indices0 = torch.where(valid0, indices0, indices0.new_tensor(-1))
        indices1 = torch.where(valid1, indices1, indices1.new_tensor(-1))

        return indices0, indices1, mscores0, mscores1