Some operations seem difficult to follow about cmrnet

[cattaneod]

rotate_back and rotate_forward are basically the same, but with the RT inverted.

In RTs I save the results:
RTs[0] is the initial estimation H_init
RTs[1] is H_init @ prediction1 (prediction of the first network)
RTs[2] is H_init @ prediction1 @ prediction2

and so on. Thus, the final RTs[-1] should be as close as possible to the identity matrix.

[RaviBeagle]

Yes. Thank you.
I do understand that rotate_back is meant to rotate the point cloud from estimated to real pose.
and rotate_forward from actual pose to erroneous pose. And only invert of RT is the difference between the two functions.

DataSetVisibility*.py
R = mathutils.Euler((rotx, roty, rotz), 'XYZ')
T = mathutils.Vector((transl_x, transl_y, transl_z))

    R, T = invert_pose(R, T)   ##  Invert Error Transform as 
    R, T = torch.tensor(R), torch.tensor(T)

But in in evaluate_*.py:

R = mathutils.Quaternion(sample['rot_error'][idx])
T = mathutils.Vector(sample['tr_error'][idx])
print(pc_rotated.shape, R, T)
pc_rotated = rotate_back(pc_rotated, R, T)

Here should we not use rotate_forward ? Or have got the understanding interchanged ?

[RaviBeagle]

I will keep the same conventions. So this is my code for now:

        max_angle = 10.0
        max_t = 2.0
        rotz = np.random.uniform(-max_angle, max_angle) * (3.141592 / 180.0)
        roty = np.random.uniform(-max_angle, max_angle) * (3.141592 / 180.0)
        rotx = np.random.uniform(-max_angle, max_angle) * (3.141592 / 180.0)
        transl_x = np.random.uniform(-max_t, max_t)
        transl_y = np.random.uniform(-max_t, max_t)
        transl_z = np.random.uniform(-max_t, min(max_t, 1.))

        R = mathutils.Euler((rotx, roty, rotz), 'XYZ')
        T = mathutils.Vector((transl_x, transl_y, transl_z))
        R_inv, T_inv = invert_pose(R, T)

        R = quat2mat(torch.tensor(R_inv,device='cuda'))
        T = tvector2mat(torch.tensor(T_inv,device='cuda'))

        RT1_inv = torch.mm(T, R)
        RT1 = RT1_inv.clone().inverse()
        rotated_point_cloud = rotate_forward(local_map, RT1)
        
        project_lidar_to_image(cam_params,rotated_point_cloud,rgb_image,True,"INPUT")