[Python] Why can't get the correct point clouds from depth map?

My depth map is generated by blender.Here are the relevant settings of blender.



This is my main scene.


Here is the depth map:


My code here:

import cv2
import numpy as np
import open3d as o3d

f_x = 1111.11
f_y = 1111.11
c_x = 400
c_y = 400

depth_map = cv2.imread("./images/depth.png", cv2.IMREAD_ANYDEPTH)
height, width = depth_map.shape

z_min = 4.0
z_max = 16.0

point_cloud = []

for v in range(height):
for u in range(width):
d_norm = depth_map[v, u] / 65535.0

z = d_norm * (z_max - z_min) + z_min
if z > 0:
X = (u - c_x) * z / f_x
Y = (v - c_y) * z / f_y

point_cloud.append(np.array([X, Y, z]))

point_cloud = np.array(point_cloud)

pcd = o3d.geometry.PointCloud()

pcd.points = o3d.utility.Vector3dVector(point_cloud)

o3d.visualization.draw_geometries([pcd])



The reconstruction is terrible, with inexplicable distortions. Camera sensor fit sets Vertical 36x36, picture resolution 800x800


It should actually be a flat surface.

Continue reading...