def vertices3d_to_box3d(vertices: Tensor):
center = vertices.sum(axis=0)
# right upper left lower in xy plane
right_index, far_index, high_index = vertices.argmax(axis=0)
left_index, _, low_index = vertices.argmin(axis=0)
x_right, y_right, _ = vertices[right_index]
x_far, y_far, _ = vertices[far_index]
x_left, y_left, _ = vertices[left_index]
_, _, z_high = vertices[high_index]
_, _, z_low = vertices[low_index]
far_right_x_delta = np.abs(x_right - x_far)
far_right_y_delta = np.abs(y_right - y_far)
far_left_x_delta = np.abs(x_left - x_far)
far_left_y_delta = np.abs(y_left - y_far)
angle = np.arctan(far_right_x_delta / far_right_y_delta) #TODO atan2
y_size = np.sqrt(far_right_x_delta**2 + far_right_y_delta**2)
x_size = np.sqrt(far_left_x_delta**2 + far_left_y_delta**2)
z_size = z_high - z_low
extent = torch.tensor([x_size, y_size, z_size])
angle = torch.tensor([angle])
box = torch.cat((center, extent, angle), dim=0)
return box
def _extremity_game_target_function(context: torch.Tensor,
function_selectors: torch.Tensor,
target_type) -> torch.Tensor:
func_idxs = function_selectors.argmax(dim=1)
func_min_or_max = func_idxs % 2
param_idxs = func_idxs // 2
min_obj_per_param = context.argmin(
dim=1) # index of min property along the row (index of the object)
max_obj_per_param = context.argmax(dim=1)
targets = []
targets2 = [] # previous targets
for batch in range(context.shape[0]):
num_object = context.size()[1]
if func_min_or_max[batch] == 0:
# batch = batch id
# min_obj_per_param[batch][param_idxs[batch]]: index of min
targets2.append(
context[batch, min_obj_per_param[batch][param_idxs[batch]]])
# MY CODE
t_id = torch.zeros(num_object).long()
o_id = min_obj_per_param[batch][param_idxs[batch]] # object id
t_id[o_id] = 1 # one-hot tensor
targets.append(o_id)
# print("min_obj_per_param[batch][param_idxs[batch]]: ", min_obj_per_param[batch][param_idxs[batch]])
else:
targets2.append(
context[batch, max_obj_per_param[batch][param_idxs[batch]]])
# MY CODE
t_id = torch.zeros(num_object).long()
o_id = max_obj_per_param[batch][param_idxs[batch]] # object id
t_id[o_id] = 1 # one-hot tensor
targets.append(o_id)
# print(" max_obj_per_param[batch][param_idxs[batch]]: ", max_obj_per_param[batch][param_idxs[batch]])
# print("context: ", context)
# print("targets: ", targets)
# print("targets2: ", targets2)
# print(" torch.stack(targets)", torch.stack(targets))
# print("torch.stack(targets2) : ", torch.stack(targets2) )
if target_type == "target_properties":
return torch.stack(targets2) #
elif target_type == "target_id": # target_id
return torch.stack(targets)
else:
print("invalid target type")
exit()
def find_bmus(self, distances: torch.Tensor) -> torch.Tensor:
"""
Find BMU for each batch in `distances`.
Parameters
----------
distances : torch.Tensor
The data-to-codebook distances.
"""
# distances shape: [bs, codebook_size]
# argmin shape : [bs, 1]
min_idxs = distances.argmin(-1)
# Distances are flattened, so we need to transform 1d indices into 2d map locations
return idxs_1d_to_2d(min_idxs, self.size[1]).to(distances.device)
def _extremity_game_target_function(
context: torch.Tensor,
function_selectors: torch.Tensor) -> torch.Tensor:
func_idxs = function_selectors.argmax(dim=1)
func_min_or_max = func_idxs % 2
param_idxs = func_idxs // 2
min_obj_per_param = context.argmin(dim=1)
max_obj_per_param = context.argmax(dim=1)
targets = []
for batch in range(context.shape[0]):
if func_min_or_max[batch] == 0:
targets.append(
context[batch, min_obj_per_param[batch][param_idxs[batch]]])
else:
targets.append(
context[batch, max_obj_per_param[batch][param_idxs[batch]]])
return torch.stack(targets)
def update(self, pixel_error: torch.Tensor):
if not self.is_fixed:
self.optimal_id = pixel_error.argmin()
