Qiangeng Xu
Qiangeng Xu
hi, it is similar to MVSNeRF, which takes 250000 iterations, almost 8 epochs. However, after 3-4 epochs, the growth is starting to be slow. it takes me about one day,...
yes, however, without initial training, you can also directly do per-scene optimization from random value, but the convergence is slower.
hi folks, it is wrong decision to build things in pycuda, and im sorry that i don't have enough time to rebuild everything in a fashion of pure cuda/pytorch.
hi, the quality of point growing from scratch is poor and slow. therefore i think it is better to be used as a modification method over inial point cloud reconstruction....
hi, im not an expert in colmap, but i think sparse recon is more for sfm and dense is for the phase of mvs.
hi, we have updated the pytorch cuda extension 5 months ago, is your extension similar?
hi we have developed everything in perspective coordinate first, so, if it is not used in the word coordinate, it might be a historical issue
hi aditya, it was developed at a very early stage (1 year and a half ago). if i can find the script, i ll submit a new commit
hi, i think you can uncomment the mesh clean up code in the script, the result is due to unclean mesh model from the original shapenet dataset.