IMC_pattern/loss.py at main · centurion-crawler/IMC_pattern · 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

import torch
import torch.nn as nn
import numpy as np
import torch.nn.functional as F

device=torch.device("cuda:0")
def lsim_loss_cal(representations,label,T=0.7): # Contrastive Learning Similarity Loss
    n = label.shape[0]

    similarity_matrix = F.cosine_similarity(representations.unsqueeze(0),representations.unsqueeze(1),dim=2)
    mask = torch.ones_like(similarity_matrix)*(label.expand(n,n).eq(label.expand(n,n).t()))

    mask_no_sim = torch.ones_like(mask)-mask

    mask_eig_0 = torch.ones(n,n)-torch.eye(n,n)
    mask_eig_0 = mask_eig_0.to(device)

    similarity_matrix = torch.exp(similarity_matrix/T)

    similarity_matrix = similarity_matrix*mask_eig_0

    sim = mask*similarity_matrix

    no_sim = similarity_matrix-sim

    no_sim_sum = torch.sum(no_sim,dim=1)

    no_sim_sum_expend = no_sim_sum.repeat(n,1).T
    sim_sum = sim + no_sim_sum_expend
    loss = torch.div(sim,sim_sum)
    loss = mask_no_sim + loss + torch.eye(n,n).to(device)

    loss = - torch.log(loss)
    loss = torch.sum(torch.sum(loss,dim=1))/(len(torch.nonzero(loss)))

    return loss