def forward(self, sample_params):
# Unpack the sample_params
class_labels = sample_params["class_labels"]
translations = sample_params["translations"]
sizes = sample_params["sizes"]
angles = sample_params["angles"]
room_layout = sample_params["room_layout"]
# shape (batch,length,dimension)
B, _, _ = class_labels.shape
# Apply the positional embeddings only on bboxes that are not the start
# token
class_f = self.fc_class(class_labels)
# Apply the positional embedding along each dimension of the position
# property
pos_f_x = self.pe_pos_x(translations[:, :, 0:1])
pos_f_y = self.pe_pos_x(translations[:, :, 1:2])
pos_f_z = self.pe_pos_x(translations[:, :, 2:3])
pos_f = torch.cat([pos_f_x, pos_f_y, pos_f_z], dim=-1)
Maybe here have some wrong order? In my dataset_stats.txt the bounds_translations "bounds_translations": [-2.762500499999998, 0.045, -2.7527500000000007, 2.778441746198965, 3.6248395981292725, 2.818542771063899],so I think maybe the z Coordinate cliped in [0.045,-3.624] ,Maybe the pos_f_y should actually be pos_f_z?
Maybe here have some wrong order? In my dataset_stats.txt the bounds_translations "bounds_translations": [-2.762500499999998, 0.045, -2.7527500000000007, 2.778441746198965, 3.6248395981292725, 2.818542771063899],so I think maybe the z Coordinate cliped in [0.045,-3.624] ,Maybe the pos_f_y should actually be pos_f_z?