def __init__(self,
path_to_theta_lut,
poly_coeffs,
principal_point=torch.Tensor([0., 0.]),
scale_factors=torch.Tensor([1., 1.]),
Tcw=None):
"""
Initializes the Camera class
Parameters
----------
intrinsics : dictionary (keys : ax, ay, cx, cy, c1, c2, c3, c5)
Camera intrinsics
poly_coeffs [c1, c2, c3, c4]
principal_point [cx, cy]
scale_factors [ax, ay]
Tcw : Pose
Camera -> World pose transformation
"""
super().__init__()
self.path_to_theta_lut = path_to_theta_lut
self.poly_coeffs = poly_coeffs
self.principal_point = principal_point
self.scale_factors = scale_factors
#self.K = K
self.Tcw = Pose.identity(
len(poly_coeffs)
) if Tcw is None else Tcw #Pose.identity(len(K)) if Tcw is None else Tcw
def __init__(self,
poly_coeffs, principal_point, scale_factors,
K, k1, k2, k3, p1, p2,
camera_type, #int Tensor ; 0 is fisheye, 1 is distorted, 2 is other
Tcw=None):
"""
Initializes the Camera class
Parameters
----------
intrinsics : dictionary (keys : ax, ay, cx, cy, c1, c2, c3, c5)
Camera intrinsics
poly_coeffs [c1, c2, c3, c4]
principal_point [cx, cy]
scale_factors [ax, ay]
Tcw : Pose
Camera -> World pose transformation
"""
super().__init__()
self.poly_coeffs = poly_coeffs
self.principal_point = principal_point
self.scale_factors = scale_factors
self.K = K
self.k1 = k1
self.k2 = k2
self.k3 = k3
self.p1 = p1
self.p2 = p2
self.camera_type = camera_type
self.Tcw = Pose.identity(len(camera_type)) if Tcw is None else Tcw
def __init__(self, K, Tcw=None):
"""
Initializes the Camera class
Parameters
----------
K : torch.Tensor [3,3]
Camera intrinsics
Tcw : Pose
Camera -> World pose transformation
"""
super().__init__()
self.K = K
self.Tcw = Pose.identity(len(K)) if Tcw is None else Tcw
def __init__(self, R, Tcw=None):
"""
Initializes the Camera class
Parameters
----------
R : torch.Tensor [B, 3, H, W]
Camera ray surface
Tcw : Pose
Camera -> World pose transformation
"""
super().__init__()
self.ray_surface = R
self.Tcw = Pose.identity(1) if Tcw is None else Tcw
import cv2
import open3d as o3d
main_folder = '/home/vbelissen/test_data/valeo_data_ready2train/data/dataset_valeo_cea_2017_2018/'
seq_idx = '20170320_144339'
img_idx = '00011702'
path_to_theta_lut = [
main_folder + 'images/fisheye/train/' + seq_idx +
'/cam_0/theta_tensor_1280_800.npy'
]
poly_coeffs = torch.Tensor([282.85, -27.8671, 114.318, -36.6703]).unsqueeze(0)
principal_point = torch.Tensor([0.046296, -7.33178]).unsqueeze(0)
scale_factors = torch.Tensor([1., 1. / 1.00173]).unsqueeze(0)
Tcw = Pose.identity(len(poly_coeffs))
Twc = Tcw.inverse()
r = R.from_quat([1, 0, 0, 0])
depth_map_valeo = np.zeros((1, 1, 800, 1280))
depth_map_valeo[0, 0, :, :] = \
np.load(main_folder + 'depth_maps/fisheye/train/' + seq_idx + '/velodyne_0/' + seq_idx + '_velodyne_0_' + img_idx + '.npz')['velodyne_depth']
depth_map_valeo = depth_map_valeo.astype('float32')
depth_map_valeo_tensor = torch.from_numpy(depth_map_valeo)
def reconstruct(depth, frame='w'):
"""
Reconstructs pixel-wise 3D points from a depth map.