def load_skeleton(self, B_path, is_clean_pose=True):
B_coor = json.load(open(B_path))["people"]
if len(B_coor) == 0:
pose = torch.zeros(self.opt.structure_nc, self.load_size[0],
self.load_size[1])
else:
B_coor = B_coor[0]
pose_dict = openpose_utils.obtain_2d_cords(
B_coor,
resize_param=self.load_size,
org_size=self.org_size,
affine=self.affine_param)
pose_body = pose_dict['body']
if not is_clean_pose:
pose_body = openpose_utils.openpose18_to_coco17(pose_body)
pose_numpy = openpose_utils.obtain_map(pose_body, self.load_size)
pose = np.transpose(pose_numpy, (2, 0, 1))
pose = torch.Tensor(pose)
Bi = pose
if not self.opt.no_bone_map:
color = np.zeros(shape=self.load_size + (3, ), dtype=np.uint8)
LIMB_SEQ = openpose_utils.LIMB_SEQ_HUMAN36M_17 if is_clean_pose else openpose_utils.LIMB_SEQ_COCO_17
color = openpose_utils.draw_joint(color,
pose_body.astype(np.int),
LIMB_SEQ)
color = np.transpose(color, (2, 0, 1))
color = torch.Tensor(color)
Bi = torch.cat((Bi, color), dim=0)
return Bi
def load_keypoint(self, B_path, is_clean_pose):
B_coor = json.load(open(B_path))["people"]
if len(B_coor)==0:
pose = torch.zeros(17*2, 1)
else:
B_coor = B_coor[0]
pose_dict = openpose_utils.obtain_2d_cords(B_coor, resize_param=self.load_size, org_size=self.org_size, affine=self.affine_param)
if not is_clean_pose:
pose = openpose_utils.openpose18_to_coco17(pose_dict['body'])
else:
pose = pose_dict['body']
pose = torch.Tensor(pose).float()
pose = pose.view(17*2, 1)
pose = 2*pose/self.load_size[1]-1
return pose
def load_keypoints(self, keypoint_list, B_paths):
pose_list=[]
B_path_recoder=[]
for index in keypoint_list:
B_path = B_paths[index]
B_path_recoder.append(B_path)
B_coor = json.load(open(B_path))["people"]
if len(B_coor)==0:
pose = torch.zeros(17*2, 1)
else:
B_coor = B_coor[0]
pose_dict = openpose_utils.obtain_2d_cords(B_coor, resize_param=self.load_size, org_size=self.org_size)
pose = openpose_utils.openpose18_to_coco17(pose_dict['body'])
pose = torch.Tensor(pose).float()
pose = pose.view(17*2, 1)
# normalize pose
pose = 2*pose/self.load_size[1]-1
pose_list.append(pose)
B = torch.cat(pose_list, 1)
return B, B_path_recoder