xucao-42 / normalintegration Goto Github PK

First of all, I would like to express my gratitude for your contribution to such a useful project. I am currently experiencing some issues with your NormalIntegration module and would appreciate your help.
I want to get the results of obj, but encountered multiple runtime warnings and errors while running the module, mainly related to invalid value calculations and file opening issues. Below are the detailed error messages and the context in which they occurred.

E:\NormalIntegration\data\data_class.py:29: RuntimeWarning: invalid value encountered in sqrt
t = (-b - np.sqrt(b ** 2 - 4 * a * c)) / (2 * a)
E:\NormalIntegration\utils.py:173: RuntimeWarning: invalid value encountered in true_divide
zx = -nx / nz
E:\NormalIntegration\utils.py:174: RuntimeWarning: invalid value encountered in true_divide
zy = -ny / nz
E:\NormalIntegration\utils.py:241: RuntimeWarning: Mean of empty slice
left_forward[..., None]), -1), -1)
E:\NormalIntegration\utils.py:244: RuntimeWarning: Mean of empty slice
bottom_forward[..., None]), -1), -1)
E:\NormalIntegration\data\data_class.py:61: RuntimeWarning: invalid value encountered in true_divide
p = - n[..., 0] / n[..., 2]
E:\NormalIntegration\data\data_class.py:62: RuntimeWarning: invalid value encountered in true_divide
q = - n[..., 1] / n[..., 2]
E:\NormalIntegration\data\data_class.py:103: RuntimeWarning: invalid value encountered in true_divide
p_map = - self.n[..., 0] / self.n[..., 2]
E:\NormalIntegration\data\data_class.py:104: RuntimeWarning: invalid value encountered in true_divide
q_map = - self.n[..., 1] / self.n[..., 2]
E:\NormalIntegration\data\data_class.py:61: RuntimeWarning: invalid value encountered in true_divide
p = - n[..., 0] / n[..., 2]
E:\NormalIntegration\data\data_class.py:62: RuntimeWarning: invalid value encountered in true_divide
q = - n[..., 1] / n[..., 2]
E:\NormalIntegration\data\data_class.py:124: RuntimeWarning: invalid value encountered in true_divide
p_map = - self.n[..., 0] / self.n[..., 2]
E:\NormalIntegration\data\data_class.py:125: RuntimeWarning: invalid value encountered in true_divide
q_map = - self.n[..., 1] / self.n[..., 2]
E:\shapeUp2014_face_quick_version\NormalIntegration\data\data_class.py:61: RuntimeWarning: invalid value encountered in true_divide
p = - n[..., 0] / n[..., 2]
E:\shapeUp2014_face_quick_version\NormalIntegration\data\data_class.py:62: RuntimeWarning: invalid value encountered in true_divide
q = - n[..., 1] / n[..., 2]
E:\NormalIntegration\data\data_sphere.py:20: RuntimeWarning: invalid value encountered in sqrt
z = np.sqrt(1 - XX ** 2 - YY ** 2)
E:\NormalIntegration\data\data_vase.py:31: RuntimeWarning: invalid value encountered in sqrt
z = np.sqrt(py ** 2 - XX ** 2)
running orthographic_five_point_plane_fitting...
running orthographic_four_point_plane_fitting...
running orthographic_poisson...
running orthographic_discrete_functional...
running orthographic_discrete_geometry_processing...
ERROR:root:Error opening PLY file
running orthographic_five_point_plane_fitting...
running orthographic_four_point_plane_fitting...
running orthographic_poisson...
running orthographic_discrete_functional...
running orthographic_discrete_geometry_processing...
ERROR:root:Error opening PLY file
ERROR:root:Error opening PLY file
ERROR:root:Error opening PLY file
ERROR:root:Error opening PLY file
ERROR:root:Error opening PLY file
ERROR:root:Error opening PLY file

Firstly, thanks for your code, I am applying your normal integration algorithm to my task, refine coarse geometry with fine normal maps

I run this script on body and cloth normal maps and reconstructed the surfaces as follows

python methods/orthographic_five_point_plane_fitting.py --path slack_trial2-000150.npy

normal map	surface-view1	surface-view2

It seems that the local offsets are correct, but the global translation is totally wrong. If I already have the coarse geometry+fine normal maps of the human, what if I want to introduce the coarse geometry into the optimization process? Should I replace the $d$ with point-to-origin distance $d_o$ computed from coarse geometry?

human.zip

npy file dict keys:

• clothed human: "normal_map_F", "normal_map_B", "mask"
• coarse body: "T_normal_F", "T_normal_B", "T_mask"

UPDATE:

OMG, I tried your ECCV 2022 work, and the results look much much much better

I cannot wait to ask what's the main difference between the Inverse Plane Fitting and Bilateral Normal Integration?

The method gives very nice results.
If I have a perspective depth map prior in camera space, how could I use it as integration boundary?