def loss(self, anchor_objectnesses: Tensor, anchor_transformers: Tensor,
gt_anchor_objectnesses: Tensor, gt_anchor_transformers: Tensor,
batch_size: int, batch_indices: Tensor) -> Tuple[Tensor, Tensor]:
cross_entropies = torch.empty(batch_size,
dtype=torch.float,
device=anchor_objectnesses.device)
# Using smooth_l1_losses so we can avoid outliers to dominant loss function
# details： https://zhuanlan.zhihu.com/p/48426076
smooth_l1_losses = torch.empty(batch_size,
dtype=torch.float,
device=anchor_transformers.device)
for batch_index in range(batch_size):
selected_indices = (
batch_indices == batch_index).nonzero().view(-1)
cross_entropy = F.cross_entropy(
input=anchor_objectnesses[selected_indices],
target=gt_anchor_objectnesses[selected_indices])
fg_indices = gt_anchor_objectnesses[selected_indices].nonzero(
).view(-1)
smooth_l1_loss = beta_smooth_l1_loss(
input=anchor_transformers[selected_indices][fg_indices],
target=gt_anchor_transformers[selected_indices][fg_indices],
beta=self._anchor_smooth_l1_loss_beta)
cross_entropies[batch_index] = cross_entropy
smooth_l1_losses[batch_index] = smooth_l1_loss
return cross_entropies, smooth_l1_losses
def loss(self, proposal_classes: Tensor, proposal_transformers: Tensor,
gt_proposal_classes: Tensor, gt_proposal_transformers: Tensor,
batch_size, batch_indices) -> Tuple[Tensor, Tensor]:
proposal_transformers = proposal_transformers.view(-1, self.num_classes, 4)[torch.arange(end=len(proposal_transformers), dtype=torch.long), gt_proposal_classes]
transformer_normalize_mean = self._transformer_normalize_mean.to(device=gt_proposal_transformers.device)
transformer_normalize_std = self._transformer_normalize_std.to(device=gt_proposal_transformers.device)
gt_proposal_transformers = (gt_proposal_transformers - transformer_normalize_mean) / transformer_normalize_std # scale up target to make regressor easier to learn
cross_entropies = torch.empty(batch_size, dtype=torch.float, device=proposal_classes.device)
smooth_l1_losses = torch.empty(batch_size, dtype=torch.float, device=proposal_transformers.device)
for batch_index in range(batch_size):
selected_indices = (batch_indices == batch_index).nonzero().view(-1)
cross_entropy = F.cross_entropy(input=proposal_classes[selected_indices],
target=gt_proposal_classes[selected_indices])
fg_indices = gt_proposal_classes[selected_indices].nonzero().view(-1)
smooth_l1_loss = beta_smooth_l1_loss(input=proposal_transformers[selected_indices][fg_indices],
target=gt_proposal_transformers[selected_indices][fg_indices],
beta=self._proposal_smooth_l1_loss_beta)
cross_entropies[batch_index] = cross_entropy
smooth_l1_losses[batch_index] = smooth_l1_loss
return cross_entropies, smooth_l1_losses
def loss(self, anchor_objectnesses: Tensor, anchor_transformers: Tensor,
gt_anchor_objectnesses: Tensor, gt_anchor_transformers: Tensor,
batch_size: int,
batch_indices: Tensor): # -> Tuple[Tensor, Tensor]:
cross_entropies = torch.empty(batch_size,
dtype=torch.float,
device=anchor_objectnesses.device)
smooth_l1_losses = torch.empty(batch_size,
dtype=torch.float,
device=anchor_transformers.device)
for batch_index in range(batch_size):
selected_indices = (
batch_indices == batch_index).nonzero().view(-1)
#This part
cross_entropy = F.cross_entropy(
input=anchor_objectnesses[selected_indices],
target=gt_anchor_objectnesses[selected_indices])
fg_indices = gt_anchor_objectnesses[selected_indices].nonzero(
).view(-1)
smooth_l1_loss = beta_smooth_l1_loss(
input=anchor_transformers[selected_indices][fg_indices],
target=gt_anchor_transformers[selected_indices][fg_indices],
beta=self._anchor_smooth_l1_loss_beta)
cross_entropies[batch_index] = cross_entropy
smooth_l1_losses[batch_index] = smooth_l1_loss
return cross_entropies, smooth_l1_losses
def loss(self, proposal_vertices: Tensor, proposal_classes: Tensor,
proposal_transformers: Tensor, gt_proposal_classes: Tensor,
gt_proposal_transformers: Tensor, gt_vertices: Tensor,
batch_size, batch_indices) -> Tuple[Tensor, Tensor]:
#proposal_transformers = proposal_transformers.view(-1, 2, 4)[torch.arange(end=len(proposal_transformers), dtype=torch.long), gt_proposal_classes]
transformer_normalize_mean = self._transformer_normalize_mean.to(
device=gt_proposal_transformers.device)
transformer_normalize_std = self._transformer_normalize_std.to(
device=gt_proposal_transformers.device)
gt_proposal_transformers = (
gt_proposal_transformers - transformer_normalize_mean
) / transformer_normalize_std # scale up target to make regressor easier to learn
cross_entropies = torch.empty(batch_size,
dtype=torch.float,
device=proposal_classes.device)
smooth_l1_losses = torch.empty(batch_size,
dtype=torch.float,
device=proposal_transformers.device)
vertex_losses = torch.empty(batch_size,
dtype=torch.float,
device=proposal_vertices.device)
bceloss = nn.BCELoss()
sigmoid = nn.Sigmoid()
for batch_index in range(batch_size):
selected_indices = (
batch_indices == batch_index).nonzero().view(-1)
# print(proposal_classes)
# print(sigmoid(proposal_classes[selected_indices]).squeeze(1))
# print(gt_proposal_classes[selected_indices].float())
cross_entropy = bceloss(
sigmoid(proposal_classes[selected_indices]).squeeze(1),
gt_proposal_classes[selected_indices].float())
fg_indices = gt_proposal_classes[selected_indices].nonzero(
).view(-1)
#print(proposal_transformers[selected_indices][fg_indices])
corner_l1_loss = beta_smooth_l1_loss(
input=proposal_transformers[selected_indices][fg_indices],
target=gt_proposal_transformers[selected_indices]
[fg_indices],
beta=self._proposal_smooth_l1_loss_beta)
vertex_loss = beta_smooth_l1_loss(
input=proposal_vertices[selected_indices][fg_indices],
target=gt_vertices[selected_indices][fg_indices].view(
-1, 16),
beta=self._proposal_smooth_l1_loss_beta)
cross_entropies[batch_index] = cross_entropy
smooth_l1_losses[batch_index] = corner_l1_loss
vertex_losses[batch_index] = vertex_loss
#print('losses', cross_entropies, smooth_l1_losses, vertex_losses)
return cross_entropies, smooth_l1_losses, vertex_losses