-
Notifications
You must be signed in to change notification settings - Fork 7
/
loss.py
157 lines (113 loc) · 4.89 KB
/
loss.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
import torch
import torch.nn as nn
from torch.nn import functional as F
from pprint import pprint
### Credits https://github.com/HobbitLong/SupContrast
class SupConLoss(nn.Module):
def __init__(self, temperature=0.07):
super(SupConLoss, self).__init__()
self.temperature = temperature
def forward(self, features, labels=None, mask=None):
device = (torch.device('cuda')
if features.is_cuda
else torch.device('cpu'))
batch_size = features.shape[0] ## 2*N
if labels is not None and mask is not None:
raise ValueError('Cannot define both `labels` and `mask`')
elif labels is None and mask is None:
mask = torch.eye(batch_size, dtype=torch.float32).to(device)
elif labels is not None:
labels = labels.contiguous().view(-1, 1)
if labels.shape[0] != batch_size:
raise ValueError('Num of labels does not match num of features')
mask = torch.eq(labels, labels.T).float().to(device)
else:
mask = mask.float().to(device)
contrast_feature = features
anchor_feature = contrast_feature
anchor_count = 2 ## we have two views
# compute logits
anchor_dot_contrast = torch.div(
torch.matmul(anchor_feature, contrast_feature.T),
self.temperature)
logits_mask = torch.scatter(
torch.ones_like(mask),
1,
torch.arange(batch_size).view(-1, 1).to(device),
0
)
## it produces 1 for the non-matching places and 0 for matching places i.e its opposite of mask
mask = mask * logits_mask
# compute log_prob with logsumexp
logits_max, _ = torch.max(anchor_dot_contrast, dim=1, keepdim=True)
logits = anchor_dot_contrast - logits_max.detach()
exp_logits = torch.exp(logits) * logits_mask
## log_prob = x - max(x1,..,xn) - logsumexp(x1,..,xn) the equation
log_prob = logits - torch.log(exp_logits.sum(1, keepdim=True))
# compute mean of log-likelihood over positive
mean_log_prob_pos = (mask * log_prob).sum(1) / mask.sum(1)
# loss
loss = -1 * mean_log_prob_pos
loss = loss.mean()
return loss
class LCL(nn.Module):
def __init__(self, temperature=0.07):
super(LCL, self).__init__()
self.temperature = temperature
def forward(self, features, labels=None, weights=None,mask=None):
"""
Returns:
A loss scalar.
"""
device = (torch.device('cuda')
if features.is_cuda
else torch.device('cpu'))
batch_size = features.shape[0]
weights = F.softmax(weights,dim=1)
if labels is not None and mask is not None:
raise ValueError('Cannot define both `labels` and `mask`')
elif labels is None and mask is None:
mask = torch.eye(batch_size, dtype=torch.float32).to(device)
elif labels is not None:
labels = labels.contiguous().view(-1, 1)
if labels.shape[0] != batch_size:
raise ValueError('Num of labels does not match num of features')
mask = torch.eq(labels, labels.T).float().to(device)
else:
mask = mask.float().to(device)
contrast_feature = features
anchor_feature = contrast_feature
anchor_count = 2
# compute logits
anchor_dot_contrast = torch.div(
torch.matmul(anchor_feature, contrast_feature.T),
self.temperature)
logits_mask = torch.scatter(
torch.ones_like(mask),
1,
torch.arange(batch_size).view(-1, 1).to(device),
0
)
## it produces 0 for the non-matching places and 1 for matching places and neg mask does the opposite
mask = mask * logits_mask
weighted_mask = torch.zeros_like(logits_mask).float().to(device)
for i,val in enumerate(labels):
for j,jval in enumerate(labels):
weighted_mask[i,j] = weights[i,jval]
weighted_mask = weighted_mask * logits_mask
pos_weighted_mask = weighted_mask * mask
# compute log_prob with logsumexp
logits_max, _ = torch.max(anchor_dot_contrast, dim=1, keepdim=True)
logits = anchor_dot_contrast - logits_max.detach()
# print(logits)
exp_logits = torch.exp(logits) * weighted_mask
## log_prob = x - max(x1,..,xn) - logsumexp(x1,..,xn) the equation
log_prob = logits - torch.log(exp_logits.sum(1, keepdim=True))
# compute mean of log-likelihood over positive
mean_log_prob_pos = (pos_weighted_mask * log_prob).sum(1) / mask.sum(1)
# loss
loss = -1 * mean_log_prob_pos
# loss = loss.view(anchor_count, batch_size).mean()
loss = loss.mean()
# print(loss)
return loss