text
stringlengths
1
93.6k
if not osp.isdir(self.args.save_inversion_path):
os.mkdir(self.args.save_inversion_path)
x_np = x[0].detach().cpu().numpy()
if ith == -1:
basename = str(self.pcd_id)
else:
basename = str(self.pcd_id)+'_'+str(ith)
np.savetxt(osp.join(self.args.save_inversion_path,basename+"_%.4d"%curr_step+'_x.txt'), x_np, fmt = "%f;%f;%f")
# save checkpoint for each stage
#self.checkpoint_flags.append('s_'+str(stage)+' x')
#self.checkpoint_pcd.append(x)
#self.checkpoint_flags.append('s_'+str(stage)+' x_map')
#self.checkpoint_pcd.append(x_map)
### save point clouds
self.x = x
if not osp.isdir(self.args.save_inversion_path):
os.mkdir(self.args.save_inversion_path)
x_np = x[0].detach().cpu().numpy()
#x_map_np = x_map[0].detach().cpu().numpy()
partial_np = self.partial[0].detach().cpu().numpy()
if ith == -1:
basename = str(self.pcd_id)
else:
basename = str(self.pcd_id)+'_'+str(ith)
if self.gt is not None:
gt_np = self.gt[0].detach().cpu().numpy()
np.savetxt(osp.join(self.args.save_inversion_path,basename+'_gt.txt'), gt_np, fmt = "%f;%f;%f")
np.savetxt(osp.join(self.args.save_inversion_path,basename+'_x.txt'), x_np, fmt = "%f;%f;%f")
#np.savetxt(osp.join(self.args.save_inversion_path,basename+'_xmap.txt'), x_map_np, fmt = "%f;%f;%f")
np.savetxt(osp.join(self.args.save_inversion_path,basename+'_partial.txt'), partial_np, fmt = "%f;%f;%f")
if save_curve:
cd_curve_arr = np.array(cd_curve_list)
np.save(osp.join(self.args.save_inversion_path,basename+'_cd_curve.npy'),cd_curve_arr)
if bool_gt:
return ucd_loss.item(), directed_hausdorff_loss.item(), gt_cd_loss.item()
else:
return ucd_loss.item(), directed_hausdorff_loss.item()
def reset_G_tmp(self):
"""
to save the pc for visualizaton
"""
self.checkpoint_pcd = [] # to save the staged checkpoints
self.checkpoint_flags = []
if self.mask_type == 'voxel_mask':
self.to_reset_mask = True # reset hole center for each shape
def reset_whole_network(self, load_path_name):
checkpoint = torch.load(load_path_name, map_location=self.args.device)
self.Encoder.load_state_dict(checkpoint['Encoder_state_dict'])
self.Decoder.load_state_dict(checkpoint['Decoder_state_dict'])
self.DI_Disentangler.load_state_dict(checkpoint['DI'])
self.DS_Disentangler.load_state_dict(checkpoint['DS'])
self.MS_Disentangler.load_state_dict(checkpoint['MS'])
#self.Z_Mapper.load_state_dict(checkpoint['Mapper'])
self.DI_Classifier.load_state_dict(checkpoint['DIC'])
self.DS_Classifier.load_state_dict(checkpoint['DSC'])
return
def set_virtual_real(self, virtual_partial=None, real_partial=None, rotmat=None, azel=None):
if virtual_partial is not None:
self.virtual_partial = virtual_partial
if real_partial is not None:
self.real_partial = real_partial
if azel is not None:
self.azel = azel
if rotmat is not None:
self.rotmat = rotmat
def set_target(self, gt=None, partial=None):
'''
set partial and gt
'''
if gt is not None:
if len(gt.shape) == 2:
self.gt = gt.unsqueeze(0)
else:
self.gt = gt
# for visualization
self.checkpoint_flags.append('GT')
self.checkpoint_pcd.append(self.gt)
else:
self.gt = None
if partial is not None:
if self.args.target_downsample_method.lower() == 'fps':
partial_size = self.args.target_downsample_size
if len(partial.shape)==2:
self.partial = self.downsample(partial.unsqueeze(0), partial_size)
else:
self.partial = self.downsample(partial, partial_size)
else:
if len(partial.shape)==2:
self.partial = partial.unsqueeze(0)
else:
self.partial = partial
else: