consistent space

Author: mingukkang

src/utils/losses.py

@@ -117,6 +117,7 @@ def __init__(self, device, batch_size, pos_collected_numerator):
     def _calculate_similarity_matrix(self):
         return self._cosine_simililarity_matrix
 
+
     def remove_diag(self, M):
         h, w = M.shape
         assert h==w, "h and w should be same"
@@ -125,10 +126,12 @@ def remove_diag(self, M):
         mask = (mask).type(torch.bool).to(self.device)
         return M[mask].view(h, -1)
 
+
     def _cosine_simililarity_matrix(self, x, y):
         v = self.cosine_similarity(x.unsqueeze(1), y.unsqueeze(0))
         return v
 
+
     def forward(self, inst_embed, proxy, negative_mask, labels, temperature, margin):
         similarity_matrix = self.calculate_similarity_matrix(inst_embed, inst_embed)
         instance_zone = torch.exp((self.remove_diag(similarity_matrix) - margin)/temperature)
@@ -161,6 +164,7 @@ def __init__(self, device, batch_size, pos_collected_numerator):
     def _calculate_similarity_matrix(self):
         return self._cosine_simililarity_matrix
 
+
     def remove_diag(self, M):
         h, w = M.shape
         assert h==w, "h and w should be same"
@@ -169,10 +173,12 @@ def remove_diag(self, M):
         mask = (mask).type(torch.bool).to(self.device)
         return M[mask].view(h, -1)
 
+
     def _cosine_simililarity_matrix(self, x, y):
         v = self.cosine_similarity(x.unsqueeze(1), y.unsqueeze(0))
         return v
 
+
     def forward(self, inst_embed, proxy, negative_mask, labels, temperature, margin):
         p2i_similarity_matrix = self.calculate_similarity_matrix(proxy, inst_embed)
         i2i_similarity_matrix = self.calculate_similarity_matrix(inst_embed, inst_embed)
@@ -202,13 +208,15 @@ def __init__(self, device, embedding_layer, num_classes, batch_size):
         self.batch_size = batch_size
         self.cosine_similarity = torch.nn.CosineSimilarity(dim=-1)
 
+
     def _get_positive_proxy_mask(self, labels):
         labels = labels.detach().cpu().numpy()
         rvs_one_hot_target = np.ones([self.num_classes, self.num_classes]) - np.eye(self.num_classes)
         rvs_one_hot_target = rvs_one_hot_target[labels]
         mask = torch.from_numpy((rvs_one_hot_target)).type(torch.bool)
         return mask.to(self.device)
 
+
     def forward(self, inst_embed, proxy, labels):
         all_labels = torch.tensor([c for c in range(self.num_classes)]).type(torch.long).to(self.device)
         positive_proxy_mask = self._get_positive_proxy_mask(labels)
@@ -231,13 +239,15 @@ def __init__(self, device, batch_size, use_cosine_similarity=True):
         self.similarity_function = self._get_similarity_function(use_cosine_similarity)
         self.criterion = torch.nn.CrossEntropyLoss(reduction="sum")
 
+
     def _get_similarity_function(self, use_cosine_similarity):
         if use_cosine_similarity:
             self._cosine_similarity = torch.nn.CosineSimilarity(dim=-1)
             return self._cosine_simililarity
         else:
             return self._dot_simililarity
 
+
     def _get_correlated_mask(self):
         diag = np.eye(2 * self.batch_size)
         l1 = np.eye((2 * self.batch_size), 2 * self.batch_size, k=-self.batch_size)
@@ -246,6 +256,7 @@ def _get_correlated_mask(self):
         mask = (1 - mask).type(torch.bool)
         return mask.to(self.device)
 
+
     @staticmethod
     def _dot_simililarity(x, y):
         v = torch.tensordot(x.unsqueeze(1), y.T.unsqueeze(0), dims=2)
@@ -254,13 +265,15 @@ def _dot_simililarity(x, y):
         # v shape: (N, 2N)
         return v
 
+
     def _cosine_simililarity(self, x, y):
         # x shape: (N, 1, C)
         # y shape: (1, 2N, C)
         # v shape: (N, 2N)
         v = self._cosine_similarity(x.unsqueeze(1), y.unsqueeze(0))
         return v
 
+
     def forward(self, zis, zjs, temperature):
         representations = torch.cat([zjs, zis], dim=0)