nerfacc
nerfacc copied to clipboard
render_transmittance_from_density crash since nerfacc.cuda.is_cub_available() is False
When I install nerfacc with torch2.0.1, render_transmittance_from_density crashed. When I read the code, I find that it is caused by the implemention of render_transmittance_from_density, also listed bellow:
def render_transmittance_from_density( t_starts: Tensor, t_ends: Tensor, sigmas: Tensor, packed_info: Optional[Tensor] = None, ray_indices: Optional[Tensor] = None, n_rays: Optional[int] = None, prefix_trans: Optional[Tensor] = None, ) -> Tuple[Tensor, Tensor]: if not is_cub_available() and packed_info is None: # Convert ray indices to packed info packed_info = pack_info(ray_indices, n_rays) ray_indices = None sigmas_dt = sigmas * (t_ends - t_starts) alphas = 1.0 - torch.exp(-sigmas_dt) trans = torch.exp( -exclusive_sum(sigmas_dt, packed_info=packed_info, indices=ray_indices) ) if prefix_trans is not None: trans = trans * prefix_trans return trans, alphas
When I input the required arguments t_starts, t_ends, sigmas; However the code step into the the packed_info = pack_info(ray_indices, n_rays)
, while ray_indices is None and error occurs.
It seems that the latest code has some bug? `
Could you provide a minimal script that reproduce it? (I.E. what's the inputs you pass into this function?)