def compute_loss(self, outputs1, outputs2, gt_points, normals, img=None):
r''' Returns the complete loss.
The full loss is adopted from the authors' implementation and
consists of
a.) Chamfer distance loss
b.) edge length loss
c.) normal loss
d.) Laplacian loss
e.) move loss
Arguments:
outputs1 (list): first outputs of model
outputs2 (list): second outputs of model
gt_points (tensor): ground truth point cloud locations
normals (tensor): normals of the ground truth point cloud
img (tensor): input images
'''
pred_vertices_1, pred_vertices_2, pred_vertices_3 = outputs1
# Chamfer Distance Loss
lc11, lc12, id11, id12 = chamfer_distance(pred_vertices_1,
gt_points,
give_id=True)
lc21, lc22, id21, id22 = chamfer_distance(pred_vertices_2,
gt_points,
give_id=True)
lc31, lc32, id31, id32 = chamfer_distance(pred_vertices_3,
gt_points,
give_id=True)
l_c = lc11.mean() + lc21.mean() + lc31.mean()
l_c2 = lc12.mean() + lc22.mean() + lc32.mean()
l_c = (l_c2 + self.param_chamfer_rel * l_c) * self.param_chamfer_w
# Edge Length Loss
l_e = (self.edge_length_loss(pred_vertices_1, 1) +
self.edge_length_loss(pred_vertices_2, 2) +
self.edge_length_loss(pred_vertices_3, 3)) * self.param_edge
# Normal Loss
l_n = (
self.normal_loss(pred_vertices_1, normals, id11, 1) +
self.normal_loss(pred_vertices_2, normals, id21, 2) +
self.normal_loss(pred_vertices_3, normals, id31, 3)) * self.param_n
# Laplacian Loss and move loss
l_l1, _ = self.laplacian_loss(pred_vertices_1, outputs2[0], block_id=1)
l_l2, move_loss1 = self.laplacian_loss(pred_vertices_2,
outputs2[1],
block_id=2)
l_l3, move_loss2 = self.laplacian_loss(pred_vertices_3,
outputs2[2],
block_id=3)
l_l = (self.param_lap_rel * l_l1 + l_l2 + l_l3) * self.param_lap
l_m = (move_loss1 + move_loss2) * self.param_move
# Final loss
loss = l_c + l_e + l_n + l_l + l_m
return loss
def eval_step(self, data):
r''' Performs an evaluation step.
The chamfer loss is calculated and returned in a dictionary.
Args:
data (tensor): input data
'''
self.model.eval()
device = self.device
points = data.get('pointcloud_chamfer').to(device)
inputs = data.get('inputs').to(device)
with torch.no_grad():
points_out = self.model(inputs)
loss = chamfer_distance(points, points_out).mean()
loss = loss.item()
eval_dict = {
'loss': loss,
'chamfer': loss,
}
return eval_dict
def eval_step(self, data):
r''' Performs an evaluation step.
The chamfer loss is calculated and returned in a dictionary.
Args:
data (tensor): input data
'''
self.model.eval()
device = self.device
evaluator = MeshEvaluator(n_points=100000)
points = data.get('pointcloud_chamfer').to(device)
inputs = data.get('inputs').to(device)
with torch.no_grad():
points_out, _ = self.model(inputs)
batch_size = points.shape[0]
points_np = points.cpu()
points_out_np = points_out.cpu().numpy()
loss = chamfer_distance(points, points_out).mean()
loss = loss.item()
eval_dict = {
'loss': loss,
}
return eval_dict
def compute_loss(self, points, inputs):
r''' Computes the loss.
The Point Set Generation Network is trained on the Chamfer distance.
Args:
points (tensor): GT point cloud data
inputs (tensor): input data for the model
'''
points_out = self.model(inputs)
loss = chamfer_distance(points, points_out).mean()
return loss
def eval_step(self, data):
r''' Performs an evaluation step.
Arguments:
data (tensor): input data
'''
self.model.eval()
points = data.get('pointcloud').to(self.device)
img = data.get('inputs').to(self.device)
normals = data.get('pointcloud.normals').to(self.device)
# Transform GT points to camera coordinates
camera_args = common.get_camera_args(data,
'pointcloud.loc',
'pointcloud.scale',
device=self.device)
world_mat, camera_mat = camera_args['Rt'], camera_args['K']
points_transformed = common.transform_points(points, world_mat)
# Transform GT normals to camera coordinates
world_normal_mat = world_mat[:, :, :3]
normals = common.transform_points(normals, world_normal_mat)
with torch.no_grad():
outputs1, outputs2 = self.model(img, camera_mat)
pred_vertices_1, pred_vertices_2, pred_vertices_3 = outputs1
loss = self.compute_loss(outputs1, outputs2, points_transformed,
normals, img)
lc1, lc2, id31, id32 = chamfer_distance(pred_vertices_3,
points_transformed,
give_id=True)
l_c = (lc1 + lc2).mean()
l_e = self.edge_length_loss(pred_vertices_3, 3)
l_n = self.normal_loss(pred_vertices_3, normals, id31, 3)
l_l, move_loss = self.laplacian_loss(pred_vertices_3,
outputs2[2],
block_id=3)
eval_dict = {
'loss': loss.item(),
'chamfer': l_c.item(),
'edge': l_e.item(),
'normal': l_n.item(),
'laplace': l_l.item(),
'move': move_loss.item()
}
return eval_dict
def compute_loss(self, points, inputs):
r''' Computes the loss.
The Point Set Generation Network is trained on the Chamfer distance.
Args:
points (tensor): GT point cloud data
inputs (tensor): input data for the model
'''
batch_size, n_steps, n_pts, dim = points.shape
points_out = self.model(inputs)
if self.loss_corr:
n_pts_pred = points_out.shape[2]
points_out = points_out.transpose(1, 2).contiguous().view(
batch_size, n_pts_pred, -1)
points = points.transpose(1, 2).contiguous().view(
batch_size, n_pts, -1)
else:
points_out = points_out.contiguous().view(batch_size * n_steps, -1,
dim)
points = points.contiguous().view(-1, n_pts, dim)
loss = chamfer_distance(points, points_out).mean()
return loss
def compute_loss(self, data):
''' Computes the loss.
Args:
data (dict): data dictionary
'''
device = self.device
encoder_inputs, raw_data = compose_inputs(
data,
mode='train',
device=self.device,
input_type=self.input_type,
depth_pointcloud_transfer=self.depth_pointcloud_transfer,
)
world_mat = None
if (self.model.encoder_world_mat is not None) \
or self.gt_pointcloud_transfer in ('view', 'view_scale_model'):
if 'world_mat' in raw_data:
world_mat = raw_data['world_mat']
else:
world_mat = get_world_mat(data, device=device)
gt_pc = compose_pointcloud(data,
device,
self.gt_pointcloud_transfer,
world_mat=world_mat)
if self.model.encoder_world_mat is not None:
out = self.model(encoder_inputs, world_mat=world_mat)
else:
out = self.model(encoder_inputs)
loss = 0
if isinstance(out, tuple):
out, trans_feat = out
if isinstance(self.model.encoder, PointNetEncoder) or isinstance(
self.model.encoder, PointNetResEncoder):
loss = loss + 0.001 * feature_transform_reguliarzer(trans_feat)
# chamfer distance loss
if self.loss_type == 'cd':
loss = loss + chamfer_distance(out, gt_pc).mean()
else:
out_pts_count = out.size(1)
loss = loss + (emd.earth_mover_distance(
out, gt_pc, transpose=False) / out_pts_count).mean()
# view penalty loss
if self.view_penalty:
gt_mask_flow = data.get('inputs.mask_flow').to(
device) # B * 1 * H * W
if world_mat is None:
world_mat = get_world_mat(data, device=device)
camera_mat = get_camera_mat(data, device=device)
# projection use world mat & camera mat
if self.gt_pointcloud_transfer == 'world_scale_model':
out_pts = transform_points(out, world_mat)
elif self.gt_pointcloud_transfer == 'view_scale_model':
t = world_mat[:, :, 3:]
out_pts = out_pts + t
elif self.gt_pointcloud_transfer == 'view':
t = world_mat[:, :, 3:]
out_pts = out_pts * t[:, 2:, :]
out_pts = out_pts + t
else:
raise NotImplementedError
out_pts_img = project_to_camera(out_pts, camera_mat)
out_pts_img = out_pts_img.unsqueeze(1) # B * 1 * n_pts * 2
out_mask_flow = F.grid_sample(gt_mask_flow,
out_pts_img) # B * 1 * 1 * n_pts
loss_mask_flow = F.relu(1. - out_mask_flow, inplace=True).mean()
loss = loss + self.loss_mask_flow_ratio * loss_mask_flow
if self.view_penalty == 'mask_flow_and_depth':
# depth test loss
t_scale = world_mat[:, 2, 3].view(world_mat.size(0), 1, 1, 1)
gt_mask = data.get('inputs.mask').byte().to(device)
depth_pred = data.get('inputs.depth_pred').to(
device) * t_scale # absolute depth from view
background_setting(depth_pred, gt_mask)
depth_z = out_pts[:, :, 2:].transpose(1, 2)
corresponding_z = F.grid_sample(
depth_pred, out_pts_img) # B * 1 * 1 * n_pts
corresponding_z = corresponding_z.squeeze(1)
# eps
loss_depth_test = F.relu(depth_z - self.depth_test_eps -
corresponding_z,
inplace=True).mean()
loss = loss + self.loss_depth_test_ratio * loss_depth_test
return loss
def eval_step(self, data):
''' Performs an evaluation step.
Args:
data (dict): data dictionary
'''
self.model.eval()
device = self.device
encoder_inputs, raw_data = compose_inputs(
data,
mode='train',
device=self.device,
input_type=self.input_type,
depth_pointcloud_transfer=self.depth_pointcloud_transfer)
world_mat = None
if (self.model.encoder_world_mat is not None) \
or self.gt_pointcloud_transfer in ('view', 'view_scale_model'):
if 'world_mat' in raw_data:
world_mat = raw_data['world_mat']
else:
world_mat = get_world_mat(data, device=device)
gt_pc = compose_pointcloud(data,
device,
self.gt_pointcloud_transfer,
world_mat=world_mat)
batch_size = gt_pc.size(0)
with torch.no_grad():
if self.model.encoder_world_mat is not None:
out = self.model(encoder_inputs, world_mat=world_mat)
else:
out = self.model(encoder_inputs)
if isinstance(out, tuple):
out, trans_feat = out
eval_dict = {}
if batch_size == 1:
pointcloud_hat = out.cpu().squeeze(0).numpy()
pointcloud_gt = gt_pc.cpu().squeeze(0).numpy()
eval_dict = self.mesh_evaluator.eval_pointcloud(
pointcloud_hat, pointcloud_gt)
# chamfer distance loss
if self.loss_type == 'cd':
loss = chamfer_distance(out, gt_pc)
else:
loss = emd.earth_mover_distance(out, gt_pc, transpose=False)
if self.gt_pointcloud_transfer in ('world_scale_model',
'view_scale_model', 'view'):
pointcloud_scale = data.get('pointcloud.scale').to(
device).view(batch_size, 1, 1)
loss = loss / (pointcloud_scale**2)
if self.gt_pointcloud_transfer == 'view':
if world_mat is None:
world_mat = get_world_mat(data, device=device)
t_scale = world_mat[:, 2:, 3:]
loss = loss * (t_scale**2)
if self.loss_type == 'cd':
loss = loss.mean()
eval_dict['chamfer'] = loss.item()
else:
out_pts_count = out.size(1)
loss = (loss / out_pts_count).mean()
eval_dict['emd'] = loss.item()
# view penalty loss
if self.view_penalty:
gt_mask_flow = data.get('inputs.mask_flow').to(
device) # B * 1 * H * W
if world_mat is None:
world_mat = get_world_mat(data, device=device)
camera_mat = get_camera_mat(data, device=device)
# projection use world mat & camera mat
if self.gt_pointcloud_transfer == 'world_scale_model':
out_pts = transform_points(out, world_mat)
elif self.gt_pointcloud_transfer == 'view_scale_model':
t = world_mat[:, :, 3:]
out_pts = out_pts + t
elif self.gt_pointcloud_transfer == 'view':
t = world_mat[:, :, 3:]
out_pts = out_pts * t[:, 2:, :]
out_pts = out_pts + t
else:
raise NotImplementedError
out_pts_img = project_to_camera(out_pts, camera_mat)
out_pts_img = out_pts_img.unsqueeze(1) # B * 1 * n_pts * 2
out_mask_flow = F.grid_sample(gt_mask_flow,
out_pts_img) # B * 1 * 1 * n_pts
loss_mask_flow = F.relu(1. - out_mask_flow,
inplace=True).mean()
loss = self.loss_mask_flow_ratio * loss_mask_flow
eval_dict['loss_mask_flow'] = loss.item()
if self.view_penalty == 'mask_flow_and_depth':
# depth test loss
t_scale = world_mat[:, 2, 3].view(world_mat.size(0), 1, 1,
1)
gt_mask = data.get('inputs.mask').byte().to(device)
depth_pred = data.get('inputs.depth_pred').to(
device) * t_scale
background_setting(depth_pred, gt_mask)
depth_z = out_pts[:, :, 2:].transpose(1, 2)
corresponding_z = F.grid_sample(
depth_pred, out_pts_img) # B * 1 * 1 * n_pts
corresponding_z = corresponding_z.squeeze(1)
# eps = 0.05
loss_depth_test = F.relu(depth_z - self.depth_test_eps -
corresponding_z,
inplace=True).mean()
loss = self.loss_depth_test_ratio * loss_depth_test
eval_dict['loss_depth_test'] = loss
return eval_dict
def eval_step_full(self, data):
r''' Performs an evaluation step.
The chamfer loss is calculated and returned in a dictionary.
Args:
data (tensor): input data
'''
self.model.eval()
device = self.device
evaluator = MeshEvaluator(n_points=100000)
points = data.get('pointcloud_chamfer').to(device)
normals = data.get('pointcloud_chamfer.normals')
inputs = data.get('inputs').to(device)
with torch.no_grad():
points_out, _ = self.model(inputs)
batch_size = points.shape[0]
points_np = points.cpu()
points_out_np = points_out.cpu().numpy()
completeness_list = []
accuracy_list = []
completeness2_list = []
accuracy2_list = []
chamfer_L1_list = []
chamfer_L2_list = []
edge_chamferL1_list = []
edge_chamferL2_list = []
emd_list = []
fscore_list = []
for idx in range(batch_size):
eval_dict_pcl = evaluator.eval_pointcloud(points_out_np[idx],
points_np[idx],
normals_tgt=normals[idx])
completeness_list.append(eval_dict_pcl['completeness'])
accuracy_list.append(eval_dict_pcl['accuracy'])
completeness2_list.append(eval_dict_pcl['completeness2'])
accuracy2_list.append(eval_dict_pcl['accuracy2'])
chamfer_L1_list.append(eval_dict_pcl['chamfer-L1'])
chamfer_L2_list.append(eval_dict_pcl['chamfer-L2'])
edge_chamferL1_list.append(eval_dict_pcl['edge-chamfer-L1'])
edge_chamferL2_list.append(eval_dict_pcl['edge-chamfer-L2'])
emd_list.append(eval_dict_pcl['emd'])
fscore_list.append(eval_dict_pcl['fscore'])
completeness = mean(completeness_list).item()
accuracy = mean(accuracy_list).item()
completeness2 = mean(completeness2_list).item()
accuracy2 = mean(accuracy2_list).item()
chamfer_L1 = mean(chamfer_L1_list).item()
chamfer_L2 = mean(chamfer_L2_list).item()
edge_chamferL1 = mean(edge_chamferL1_list).item()
edge_chamferL2 = mean(edge_chamferL2_list).item()
emd = mean(emd_list).item()
fscore = mean(fscore_list)
loss = chamfer_distance(points, points_out).mean()
loss = loss.item()
eval_dict = {
'loss': loss,
'chamfer': loss,
'completeness': completeness,
'completeness2': completeness2,
'accuracy': accuracy,
'accuracy2': accuracy2,
'chamfer-L1': chamfer_L1,
'chamfer-L2': chamfer_L2,
'edge-chamfer-L2': edge_chamferL2,
'edge-chamfer-L1': edge_chamferL1,
'emd': emd,
'fscore': fscore
}
return eval_dict
def plot_dot_step(self, data, epoch, m):
self.model.train()
anc_points = data.get('pointcloud').to(self.device)
anc_inputs = data.get('inputs').to(self.device)
pos_inputs = data.get('inputs.Bias').to(self.device)
anc_points_out, anc_feature = self.model(anc_inputs)
anc_loss = chamfer_distance(anc_points, anc_points_out).mean()
_, pos_feature = self.model(pos_inputs)
batch_size = anc_feature.shape[0]
pdist = nn.PairwiseDistance(p=2)
anc_pos_dist = pdist(anc_feature, pos_feature)
anc_pos_dot = []
for m in range(batch_size):
anc_pos = torch.dot(anc_feature[m], pos_feature[m])
anc_pos_dot.append(anc_pos)
# import pudb; pu.db
anc_pos_cham = chamfer_distance(anc_points, anc_points)
anc_neg_dist_list_pos = []
mix_dist_list_pos = []
anc_neg_cham_list_pos = []
mix_cham_list_pos = []
# import pudb; pu.db
for n in range(batch_size):
anc_list = torch.stack([anc_feature[n]] * (2 * batch_size - 2),
dim=0)
neg_list = torch.cat([
anc_feature[0:n], anc_feature[n + 1:], pos_feature[0:n],
pos_feature[n + 1:]
],
dim=0)
anc_neg_dist = pdist(anc_list, neg_list)
neg_indexes = [
idx for idx, dist in enumerate(anc_neg_dist)
if (anc_pos_dist[n] < dist) and (dist < anc_pos_dist[n] + m)
]
# neg_num = len(neg_indexes)
anc_neg_dot = []
for m in range(2 * batch_size - 2):
anc_neg = torch.dot(anc_list[m], neg_list[m])
anc_neg_dot.append(anc_neg)
anc_neg_list = torch.Tensor(anc_neg_dot)[neg_indexes]
anc_neg_dist_list_pos.extend(anc_neg_list)
mix_dist_list_pos.extend(torch.Tensor(anc_neg_dot))
anc_points_list = torch.stack([anc_points[n]] *
(2 * batch_size - 2),
dim=0)
neg_points = torch.cat([
anc_points[0:n], anc_points[n + 1:], anc_points[0:n],
anc_points[n + 1:]
],
dim=0)
# neg_points_list = neg_points[neg_indexes]
anc_mix_cham = chamfer_distance(anc_points_list, neg_points)
anc_neg_cham = anc_mix_cham[neg_indexes]
anc_neg_cham_list_pos.extend(anc_neg_cham)
mix_cham_list_pos.extend(anc_mix_cham)
anc_neg_dist_list = []
mix_dist_list = []
anc_neg_cham_list = []
mix_cham_list = []
for n in range(batch_size):
anc_list = torch.stack([anc_feature[n]] * (batch_size - 1), dim=0)
neg_list = torch.cat([anc_feature[0:n], anc_feature[n + 1:]],
dim=0)
anc_neg_dist = pdist(anc_list, neg_list)
neg_indexes = [
idx for idx, dist in enumerate(anc_neg_dist)
if (anc_pos_dist[n] < dist) and (dist < anc_pos_dist[n] + m)
]
# neg_num = len(neg_indexes)
# anc_neg_list = anc_neg_dist[neg_indexes]
anc_neg_dot = []
for m in range(batch_size - 1):
anc_neg = torch.dot(anc_list[m], neg_list[m])
anc_neg_dot.append(anc_neg)
anc_neg_list = torch.Tensor(anc_neg_dot)[neg_indexes]
anc_neg_dist_list.extend(anc_neg_list)
mix_dist_list.extend(torch.Tensor(anc_neg_dot))
anc_points_list = torch.stack([anc_points[n]] * (batch_size - 1),
dim=0)
neg_points = torch.cat([anc_points[0:n], anc_points[n + 1:]],
dim=0)
# neg_points_list = neg_points[neg_indexes]
anc_mix_cham = chamfer_distance(anc_points_list, neg_points)
anc_neg_cham = anc_mix_cham[neg_indexes]
anc_neg_cham_list.extend(anc_neg_cham)
mix_cham_list.extend(anc_mix_cham)
anc_neg_dist_list_pos = torch.Tensor(
anc_neg_dist_list_pos).cpu().detach().numpy().tolist()
mix_dist_list_pos = torch.Tensor(
mix_dist_list_pos).cpu().detach().numpy().tolist()
anc_neg_cham_list_pos = torch.Tensor(
anc_neg_cham_list_pos).cpu().detach().numpy().tolist()
mix_cham_list_pos = torch.Tensor(
mix_cham_list_pos).cpu().detach().numpy().tolist()
anc_neg_dist_list = torch.Tensor(
anc_neg_dist_list).cpu().detach().numpy().tolist()
mix_dist_list = torch.Tensor(
mix_dist_list).cpu().detach().numpy().tolist()
anc_neg_cham_list = torch.Tensor(
anc_neg_cham_list).cpu().detach().numpy().tolist()
mix_cham_list = torch.Tensor(
mix_cham_list).cpu().detach().numpy().tolist()
anc_pos_dot = torch.Tensor(anc_pos_dot).cpu().detach().numpy().tolist()
anc_pos_cham = anc_pos_cham.cpu().detach().numpy().tolist()
return anc_loss.item(
), anc_neg_dist_list_pos, mix_dist_list_pos, anc_neg_cham_list_pos, mix_cham_list_pos, anc_neg_dist_list, mix_dist_list, anc_neg_cham_list, mix_cham_list, anc_pos_dot, anc_pos_cham
def compute_loss(self, epoch, anc_points, anc_inputs, m, reg):
r''' Computes the loss.
The Point Set Generation Network is trained on the Chamfer distance.
Args:
points (tensor): GT point cloud data
inputs (tensor): input data for the model
'''
anc_points_out, anc_feature = self.model(anc_inputs)
anc_loss = chamfer_distance(anc_points, anc_points_out).mean()
# _, pos_feature = self.model(pos_inputs)
# pdist = nn.PairwiseDistance(p=2)
# anc_pos_dist = pdist(anc_feature, pos_feature)
batch_size = anc_feature.shape[0]
trip_list = []
reg_list = []
# vals = 0.11
sigma = m * 0.997 / 3
# l1 = nn.L1Loss()
# for n in range(batch_size):
# anc_list = torch.stack([anc_feature[n]]*(2*batch_size-2), dim=0)
# neg_list = torch.cat([anc_feature[0:n], anc_feature[n+1:], pos_feature[0:n], pos_feature[n+1:]], dim=0)
# anc_neg_dist = pdist(anc_list, neg_list)
# anc_neg_list = anc_neg_dist[(anc_pos_dist[n] < anc_neg_dist) * (anc_neg_dist < anc_pos_dist[n] + m)]
# neg_num = anc_neg_list.shape[0]
# if neg_num !=0:
# triploss_list = [torch.max(torch.tensor([anc_pos_dist[n] - anc_neg_list[n_neg] + m, 0])) for n_neg in range(neg_num)]
# triplet_loss = torch.mean(torch.tensor(triploss_list))
# trip_list.append(triplet_loss)
if reg == 'gau':
for n in range(batch_size):
t_feature = anc_feature[n]
t_feature_list = torch.stack([anc_feature[n]] *
(batch_size - 1),
dim=0)
s_feature_list = torch.cat(
[anc_feature[0:n], anc_feature[n + 1:]], dim=0)
t_points_list = torch.stack([anc_points[n]] * (batch_size - 1),
dim=0)
t_points = anc_points[n]
s_points_list = torch.cat(
[anc_points[0:n], anc_points[n + 1:]], dim=0)
s_num = batch_size - 1
# t_points_list = t_points[None, :, :]
# t_points_list = t_points_list.repeat(s_points_list.shape[0],1,1)
p_deno = torch.sum(
torch.exp(-1 *
chamfer_distance(t_points_list, s_points_list) /
(2 * sigma)))
p_hat_deno = torch.sum(
torch.einsum('bl, bl -> b', t_feature_list,
s_feature_list))
# p_hat_deno = sum([torch.dot(t_feature, t_feature_list[s_idx]) for s_idx in range(s_num)])
# p_hat_nume =
# t_reg_list = []
# reg_t = 0.
p = torch.exp(
-1 * chamfer_distance(t_points_list, s_points_list) /
(2 * sigma)) / p_deno
p_hat = torch.einsum('bl, bl -> b', t_feature_list,
s_feature_list) / p_hat_deno
reg_t = torch.mean(torch.abs(p_hat - p))
# torch.sum(F.l1_loss(p_hat, p))
# for s_idx in range(s_num):
# reg_t += F.l1_loss((torch.dot(t_feature, s_feature_list[s_idx]) / p_hat_deno),
# (torch.exp(-1 * chamfer_distance(t_points[None,...], s_points_list[s_idx][None,...])/ (2 * sigma)) / p_deno))
# # t_reg_list.append(reg)
# reg_t = reg_t / s_num
# t_reg = torch.mean(torch.stack(t_reg_list), dim=0)
reg_list.append(reg_t)
else:
for n in range(batch_size):
t_feature = anc_feature[n]
t_feature_list = torch.stack([anc_feature[n]] *
(batch_size - 1),
dim=0)
s_feature_list = torch.cat(
[anc_feature[0:n], anc_feature[n + 1:]], dim=0)
t_points_list = torch.stack([anc_points[n]] * (batch_size - 1),
dim=0)
# t_points = anc_points[n]
s_points_list = torch.cat(
[anc_points[0:n], anc_points[n + 1:]], dim=0)
s_num = batch_size - 1
# p_deno = sum([1 - (chamfer_distance(t_points[None,...], s_points_list[s_idx][None,...]) / vals) for s_idx in range(s_num)])
p_deno = torch.sum(
1 - chamfer_distance(t_points_list, s_points_list) / vals)
p_hat_deno = torch.sum(
torch.einsum('bl, bl -> b', t_feature_list,
s_feature_list))
# p_hat_nume =
# t_reg_list = []
p = (1 - chamfer_distance(t_points_list, s_points_list) /
vals) / p_deno
p_hat = torch.einsum('bl, bl -> b', t_feature_list,
s_feature_list) / p_hat_deno
reg_t = torch.mean(torch.abs(p_hat - p))
# reg_t = 0.
# for s_idx in range(s_num):
# reg_t += F.l1_loss((torch.dot(t_feature, s_feature_list[s_idx]) / p_hat_deno),
# (1 - (chamfer_distance(t_points[None,...], s_points_list[s_idx][None,...]) / vals) / p_deno))
# # t_reg_list.append(reg)
# reg_t = reg_t / s_num
# t_reg = torch.mean(torch.stack(t_reg_list), dim=0)
reg_list.append(reg_t)
reg_loss = torch.mean(torch.stack(reg_list), dim=0)
# triplet_output = torch.mean(torch.tensor(trip_list))
# triplet_loss = nn.TripletMarginLoss(margin=m, p=2)
# triplet_output = triplet_loss(anc_feature, pos_feature, neg_feature)
w_a = 1
# w_t = 1
w_r = 1
# loss = w_a * anc_loss + w_t * triplet_output + w_r * reg_loss
loss = w_a * anc_loss + w_r * reg_loss
return loss, anc_loss, reg_loss
