refactoring loss functions
Browse files- src/loss.py +34 -10
src/loss.py
CHANGED
@@ -3,6 +3,23 @@ from torch import nn
|
|
3 |
import torch.nn.functional as F
|
4 |
|
5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
def contrastive_loss(logits, dim):
|
7 |
neg_ce = torch.diag(F.log_softmax(logits, dim=dim))
|
8 |
return -neg_ce.mean()
|
@@ -17,25 +34,29 @@ class CLIPLoss(nn.Module):
|
|
17 |
super().__init__()
|
18 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
19 |
|
20 |
-
def forward(self,
|
21 |
temperature = self.logit_temperature.sigmoid()
|
22 |
-
|
23 |
caption_loss = contrastive_loss(similarity_matrix / temperature, dim=0)
|
24 |
image_loss = contrastive_loss(similarity_matrix / temperature, dim=1)
|
25 |
|
26 |
return 0.5 * (caption_loss + image_loss)
|
27 |
|
28 |
|
29 |
-
class
|
30 |
def __init__(self, logit_temperature: float = -1.0):
|
31 |
super().__init__()
|
32 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
33 |
self.lambda_1: float = 1.0
|
34 |
self.lambda_2: float = 1.0
|
35 |
|
36 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
37 |
temperature = self.logit_temperature.sigmoid()
|
38 |
-
similarity_matrix = image_features @ text_features.T
|
39 |
caption_loss = contrastive_loss(similarity_matrix / temperature, dim=0)
|
40 |
image_loss = contrastive_loss(similarity_matrix / temperature, dim=1)
|
41 |
|
@@ -56,9 +77,8 @@ class SigLIPLoss(nn.Module):
|
|
56 |
super().__init__()
|
57 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
58 |
|
59 |
-
def forward(self,
|
60 |
temperature = self.logit_temperature.sigmoid()
|
61 |
-
similarity_matrix = image_features @ text_features.T
|
62 |
return contrastive_sigmoid_loss(similarity_matrix / temperature)
|
63 |
|
64 |
|
@@ -69,9 +89,13 @@ class CySigLIPLoss(nn.Module):
|
|
69 |
self.lambda_1: float = 1.0
|
70 |
self.lambda_2: float = 1.0
|
71 |
|
72 |
-
def forward(
|
|
|
|
|
|
|
|
|
|
|
73 |
temperature = self.logit_temperature.sigmoid()
|
74 |
-
similarity_matrix = image_features @ text_features.T
|
75 |
loss = contrastive_sigmoid_loss(similarity_matrix / temperature)
|
76 |
|
77 |
symmetry_loss = F.mse_loss(similarity_matrix, similarity_matrix.T)
|
@@ -85,7 +109,7 @@ class CySigLIPLoss(nn.Module):
|
|
85 |
def get_loss(loss_type: str):
|
86 |
loss_functions = {
|
87 |
"clip": CLIPLoss(),
|
88 |
-
"cyclip":
|
89 |
"sigmoid": SigLIPLoss(),
|
90 |
"cyclic_sigmoid": CySigLIPLoss(),
|
91 |
}
|
|
|
3 |
import torch.nn.functional as F
|
4 |
|
5 |
|
6 |
+
def metrics(similarity: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
|
7 |
+
y = torch.arange(len(similarity)).to(similarity.device)
|
8 |
+
img2cap_match_idx = similarity.argmax(dim=1)
|
9 |
+
cap2img_match_idx = similarity.argmax(dim=0)
|
10 |
+
|
11 |
+
img_acc = (img2cap_match_idx == y).float().mean()
|
12 |
+
cap_acc = (cap2img_match_idx == y).float().mean()
|
13 |
+
|
14 |
+
return img_acc, cap_acc
|
15 |
+
|
16 |
+
|
17 |
+
def get_similarity_matrix(
|
18 |
+
image_features: torch.Tensor, text_features: torch.Tensor
|
19 |
+
) -> torch.Tensor:
|
20 |
+
return image_features @ text_features.T
|
21 |
+
|
22 |
+
|
23 |
def contrastive_loss(logits, dim):
|
24 |
neg_ce = torch.diag(F.log_softmax(logits, dim=dim))
|
25 |
return -neg_ce.mean()
|
|
|
34 |
super().__init__()
|
35 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
36 |
|
37 |
+
def forward(self, similarity_matrix: torch.Tensor):
|
38 |
temperature = self.logit_temperature.sigmoid()
|
39 |
+
|
40 |
caption_loss = contrastive_loss(similarity_matrix / temperature, dim=0)
|
41 |
image_loss = contrastive_loss(similarity_matrix / temperature, dim=1)
|
42 |
|
43 |
return 0.5 * (caption_loss + image_loss)
|
44 |
|
45 |
|
46 |
+
class CyCLIPLoss(nn.Module):
|
47 |
def __init__(self, logit_temperature: float = -1.0):
|
48 |
super().__init__()
|
49 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
50 |
self.lambda_1: float = 1.0
|
51 |
self.lambda_2: float = 1.0
|
52 |
|
53 |
+
def forward(
|
54 |
+
self,
|
55 |
+
similarity_matrix: torch.Tensor,
|
56 |
+
image_features: torch.Tensor,
|
57 |
+
text_features: torch.Tensor,
|
58 |
+
):
|
59 |
temperature = self.logit_temperature.sigmoid()
|
|
|
60 |
caption_loss = contrastive_loss(similarity_matrix / temperature, dim=0)
|
61 |
image_loss = contrastive_loss(similarity_matrix / temperature, dim=1)
|
62 |
|
|
|
77 |
super().__init__()
|
78 |
self.logit_temperature = nn.Parameter(torch.tensor(logit_temperature))
|
79 |
|
80 |
+
def forward(self, similarity_matrix: torch.Tensor):
|
81 |
temperature = self.logit_temperature.sigmoid()
|
|
|
82 |
return contrastive_sigmoid_loss(similarity_matrix / temperature)
|
83 |
|
84 |
|
|
|
89 |
self.lambda_1: float = 1.0
|
90 |
self.lambda_2: float = 1.0
|
91 |
|
92 |
+
def forward(
|
93 |
+
self,
|
94 |
+
similarity_matrix: torch.Tensor,
|
95 |
+
image_features: torch.Tensor,
|
96 |
+
text_features: torch.Tensor,
|
97 |
+
):
|
98 |
temperature = self.logit_temperature.sigmoid()
|
|
|
99 |
loss = contrastive_sigmoid_loss(similarity_matrix / temperature)
|
100 |
|
101 |
symmetry_loss = F.mse_loss(similarity_matrix, similarity_matrix.T)
|
|
|
109 |
def get_loss(loss_type: str):
|
110 |
loss_functions = {
|
111 |
"clip": CLIPLoss(),
|
112 |
+
"cyclip": CyCLIPLoss(),
|
113 |
"sigmoid": SigLIPLoss(),
|
114 |
"cyclic_sigmoid": CySigLIPLoss(),
|
115 |
}
|