src/utils/gaussian_model.py

import os
import torch
from einops import rearrange
import numpy as np
from plyfile import PlyData, PlyElement
from os import makedirs, path
from errno import EEXIST

def mkdir_p(folder_path):
    # Creates a directory. equivalent to using mkdir -p on the command line
    try:
        makedirs(folder_path)
    except OSError as exc: # Python >2.5
        if exc.errno == EEXIST and path.isdir(folder_path):
            pass
        else:
            raise

def RGB2SH(rgb):
    return (rgb - 0.5) / C0

C0 = 0.28209479177387814

# https://github.com/facebookresearch/pytorch3d/blob/main/pytorch3d/transforms/rotation_conversions.py
def quaternion_to_matrix(
    quaternions,
    eps=1e-8,
) :
    # Order changed to match scipy format!
    i, j, k, r = torch.unbind(quaternions, dim=-1)
    two_s = 2 / ((quaternions * quaternions).sum(dim=-1) + eps)

    o = torch.stack(
        (
            1 - two_s * (j * j + k * k),
            two_s * (i * j - k * r),
            two_s * (i * k + j * r),
            two_s * (i * j + k * r),
            1 - two_s * (i * i + k * k),
            two_s * (j * k - i * r),
            two_s * (i * k - j * r),
            two_s * (j * k + i * r),
            1 - two_s * (i * i + j * j),
        ),
        -1,
    )
    return rearrange(o, "... (i j) -> ... i j", i=3, j=3)


def build_covariance(
    scale,
    rotation_xyzw,
):
    scale = scale.diag_embed()
    rotation = quaternion_to_matrix(rotation_xyzw)
    return (
        rotation
        @ scale
        @ rearrange(scale, "... i j -> ... j i")
        @ rearrange(rotation, "... i j -> ... j i")
    )

def inverse_sigmoid(x):
    return torch.log(x/(1-x))

class GaussianModel:

    def __init__(self, sh_degree : int):
        self.active_sh_degree = 0
        self.max_sh_degree = sh_degree
        self._xyz = torch.empty(0)
        self._features_dc = torch.empty(0)
        self._features_rest = torch.empty(0)
        self._scaling = torch.empty(0)
        self._rotation = torch.empty(0)
        self._opacity = torch.empty(0)
        self.max_radii2D = torch.empty(0)
        self.xyz_gradient_accum = torch.empty(0)
        self.denom = torch.empty(0)
        self.optimizer = None
        self.percent_dense = 0
        self.spatial_lr_scale = 0
        self._semantic_feature = torch.empty(0) 

    @property
    def get_scaling(self):
        return self._scaling

    @property
    def get_rotation(self):
        return self._rotation

    @property
    def get_xyz(self):
        return self._xyz

    @property
    def get_features(self):
        features_dc = self._features_dc
        features_rest = self._features_rest
        return torch.cat((features_dc, features_rest), dim=1)

    @property
    def get_opacity(self):
        return self._opacity

    @property
    def get_semantic_feature(self):
        return self._semantic_feature

    def construct_list_of_attributes(self):
        l = ['x', 'y', 'z', 'nx', 'ny', 'nz']
        # All channels except the 3 DC
        for i in range(self._features_dc.shape[1]*self._features_dc.shape[2]):
            l.append('f_dc_{}'.format(i))
        for i in range(self._features_rest.shape[1]*self._features_rest.shape[2]):
            l.append('f_rest_{}'.format(i))

        l.append('opacity')
        for i in range(self._scaling.shape[1]):
            l.append('scale_{}'.format(i))
        for i in range(self._rotation.shape[1]):
            l.append('rot_{}'.format(i))
        # Add semantic features
        for i in range(self._semantic_feature.shape[1]*self._semantic_feature.shape[2]):  
            l.append('semantic_{}'.format(i))
        return l
    
    @staticmethod
    def from_predictions(pred, sh_degree):
        gaussians = GaussianModel(sh_degree=sh_degree)
        gaussians._xyz = pred['means']
        gaussians._features_dc = pred['sh_coeffs'][:, :1] # N, 1, d_sh
        gaussians._features_rest = pred['sh_coeffs'][:, 1:] # N, d_sh-1, d_sh
        gaussians._opacity = pred['opacities'] # N, 1
        gaussians._scaling = pred['scales'] # N, 3, 3
        gaussians._rotation = pred['rotations'] # N, 4
        gaussians._semantic_feature = pred['gs_feats'][:, None, :] # N, 1, d_feats
        return gaussians

    def save_ply(self, path):
        mkdir_p(os.path.dirname(path))

        xyz = self._xyz.detach().cpu().numpy()
        normals = np.zeros_like(xyz)
        f_dc = self._features_dc.detach().transpose(1, 2).flatten(start_dim=1).contiguous().cpu().numpy()
        f_rest = self._features_rest.detach().transpose(1, 2).flatten(start_dim=1).contiguous().cpu().numpy()
        opacities = inverse_sigmoid(self._opacity).detach().cpu().numpy()
        scale = torch.log(self._scaling).detach().cpu().numpy()
        rotation = self._rotation.detach().cpu().numpy()
        semantic_feature = self._semantic_feature.detach().transpose(1, 2).flatten(start_dim=1).contiguous().cpu().numpy() 

        dtype_full = [(attribute, 'f4') for attribute in self.construct_list_of_attributes()]

        elements = np.empty(xyz.shape[0], dtype=dtype_full)
        attributes = np.concatenate((xyz, normals, f_dc, f_rest, opacities, scale, rotation, semantic_feature), axis=1) 
        # attributes = np.concatenate((xyz, normals, f_dc, f_rest, opacities, scale, rotation), axis=1) 
        elements[:] = list(map(tuple, attributes))
        el = PlyElement.describe(elements, 'vertex')
        PlyData([el]).write(path)