AutoLidarPerception / SqueezeSeg

Implementation of SqueezeSeg, convolutional neural networks for LiDAR point clout segmentation https://arxiv.org/abs/1710.07368
BSD 2-Clause "Simplified" License
23 stars 7 forks source link

parameters in segment_node.py #4

Closed StevenWang30 closed 6 years ago

StevenWang30 commented 6 years ago

in line 314 in segment_node.py, what's the mean of np.radians(2.)?

    x, y, z, i = velo_points[:, 0], velo_points[:, 1], velo_points[:, 2], velo_points[:, 3]
    d = np.sqrt(x ** 2 + y ** 2 + z**2)
    r = np.sqrt(x ** 2 + y ** 2)
    d[d==0] = 0.000001
    r[r==0] = 0.000001
    phi = np.radians(45.) - np.arcsin(y/r)
    phi_ = (phi/dphi).astype(int)
    phi_[phi_<0] = 0
    phi_[phi_>=512] = 511

    # print(np.min(phi_))
    # print(np.max(phi_))
    #
    # print z
    # print np.radians(2.)
    # print min(np.arcsin(z/d))
    theta = **np.radians(2.)** - np.arcsin(z/d)
    # theta = np.radians(10.)-np.arcsin(z/d)
    # print theta
    theta_ = (theta/dtheta).astype(int)
    # print theta_
    theta_[theta_<0] = 0
    theta_[theta_>=64] = 63
Durant35 commented 6 years ago

This is following the LiDAR hardware parameters as following: image

For more information, you can refer: https://github.com/Durant35/SqueezeSeg/issues/1

Durant35 commented 6 years ago

long time no reply...