Open GaryGky opened 3 years ago
The intrinsic matrix is necessary. So you have the poses but not the intrinsic matrix? If not, you can follow the instructions from NeRF to estimate the camera poses along with the intrinsic using COLMAP, then train GRF using the default LLFF hyperparamaters. To do that, follow the section on generating your own poses from the NeRF repo at https://github.com/bmild/nerf
OK. Thank you! May I ask one more thing? I simply use this formula to generate intrinsic which assumes that the camera is without Distortion, is that reasonable?
intrinsic = np.array([[focal, 0., W / 2], [0, focal, H / 2], [0, 0, 1.]])
Seems reasonable to me, assuming the principal point isn't too far from the center.
Thank you. I will have a try.
May I ask how to get the intrinsic matrix of a photo if I want to use my own data to train GRF? And without the intrinsic matrix, will the performance of GRF significantly degrade?