yitongx / sinerf Goto Github PK

Hi, your work is very nice. It inspires me. However, I am confused about the result shown in Table 3, which indicates that either SIREN or MRS alone would negatively impact the results in the Fortress or Trex scenes. Surprisingly, when used together, they achieve the best result, surpassing even the baseline NeRF--. Can you please explain the details behind this? I would greatly appreciate it. Thanks.

Thanks for your great work!when I run your code but meet some problem.
I find in your code.you initialize pose parameters as zeros.
self.r=nn.Parameter(torch.zeros(size=(num_cams, 3), dtype=torch.float32), requires_grad=learn_R)
the rotation set to zero.but when run function Exp(r),will compute and divide norm_r.In the first iter,the norm_r will be zero.Should I change the initialize pose parameters?

def Exp(r):
    """so(3) vector to SO(3) matrix
    :param r: (3, ) axis-angle, torch tensor
    :return:  (3, 3)
    """
    skew_r = vec2skew(r)  # (3, 3)
    norm_r = r.norm() + 1e-15
    print("r:",r)
    print("norm_r:",norm_r)
    eye = torch.eye(3, dtype=torch.float32, device=r.device)
    R = eye + (torch.sin(norm_r) / norm_r) * skew_r + ((1 - torch.cos(norm_r)) / norm_r**2) * (skew_r @ skew_r)

when run the vis_learned_poses.py , it reported
ImportError: cannot import name 'get_camera_frustum_geometry_nparray' from 'utils.vis_cam_traj'

and there is no 'get_camera_frustum_geometry_nparray' function in the 'utils.vis_cam_traj' file