question about the angle representation

[Xiao-Fangff]

Excellent work! Could you please clarify whether the angle representation used in motionscript is Euler angles or rotation vectors?
I noticed in get_pose_sequence_data_from_file_HumanML3D_T2M_GPT() that the angle representation is Euler angles read from the root_euler folder. However, in generate_text_HumanML3D(), the angles are converted using rotvec_to_eulerangles, which seems to indicate that the angle representation is a rotation vector?

MotionScript/src/utils.py

Lines 913 to 928 in f7075d4

	FirstPerson_pred_xyz, pred_xyz, r_rot_quat, r_pos, euler_angles_from_quat = recover_from_ric((motion_tensor).float(), 22)
	# pred_xyz = recover_from_ric((pred_pose * std + mean).float(), 22)
	# xyz = pred_xyz.reshape(1, -1, 22, 3)
	try:
	euler_angles_from_file = np.load(npy_file_euler_address)
	euler_angles = euler_angles_from_file.squeeze()
	# euler_angles = euler_angles_from_quat
	except: # for the HumanAct12 subdataset in HumanML3D which doesn't have Euler info.
	euler_angles = np.zeros((pred_xyz.shape[0],3))
	euler_angles[:,0] = np.pi/2
	xyz = pred_xyz.reshape(pred_xyz.shape[0], -1)
	FirstPerson_xyz = FirstPerson_pred_xyz.reshape(FirstPerson_pred_xyz.shape[0], -1)
	return FirstPerson_xyz, xyz, euler_angles, r_pos

MotionScript/src/captioning_motion.py

Lines 590 to 602 in f7075d4

	for frame in range(root_euler_orient.shape[0]):
	# theta_x, theta_y, theta_z = root_orient[frame, [0, 1, 2]] # utils.rotvec_to_eulerangles(root_orient[frame, :].unsqueeze(0))
	theta_x, theta_y, theta_z = utils.rotvec_to_eulerangles(root_rot_vec[frame, :].unsqueeze(0))
	# root_orient[frame, :] = torch.cat((rad2deg(theta_x),
	# rad2deg(theta_y),
	# rad2deg(theta_z)))
	root_euler_orient[frame, :] = torch.cat(( torch.unsqueeze(rad2deg(theta_x), 0),
	torch.unsqueeze(rad2deg(theta_y), 0),
	torch.unsqueeze(rad2deg(theta_z), 0))).squeeze()