mirror of
https://github.com/graphdeco-inria/gaussian-splatting
synced 2025-04-25 00:34:39 +00:00
204 lines
9.1 KiB
Python
204 lines
9.1 KiB
Python
#
|
||
# Copyright (C) 2023, Inria
|
||
# GRAPHDECO research group, https://team.inria.fr/graphdeco
|
||
# All rights reserved.
|
||
#
|
||
# This software is free for non-commercial, research and evaluation use
|
||
# under the terms of the LICENSE.md file.
|
||
#
|
||
# For inquiries contact george.drettakis@inria.fr
|
||
#
|
||
|
||
import torch
|
||
import math
|
||
from diff_gaussian_rasterization import GaussianRasterizationSettings, GaussianRasterizer
|
||
from scene.gaussian_model import GaussianModel
|
||
from utils.sh_utils import eval_sh
|
||
from utils.general_utils import build_rotation
|
||
import torch.nn.functional as F
|
||
import matplotlib.pyplot as plt
|
||
|
||
def render(viewpoint_camera, pc : GaussianModel, pipe, bg_color : torch.Tensor, scaling_modifier = 1.0, override_color = None,
|
||
return_depth=False, return_normal=False, return_opacity=False):
|
||
"""
|
||
渲染场景： 将高斯分布的点投影到2D屏幕上来生成渲染图像
|
||
viewpoint_camera: 训练相机集合
|
||
pc: 高斯模型
|
||
pipe: 管道相关参数
|
||
bg_color: Background tensor 必须 on GPU
|
||
scaling_modifier:
|
||
override_color:
|
||
"""
|
||
|
||
# Create zero tensor. We will use it to make pytorch return gradients of the 2D (screen-space) means
|
||
# 创建一个与输入点云（高斯模型）大小相同的 零tensor，用于记录屏幕空间中的点的位置。这个张量将用于计算对于屏幕空间坐标的梯度
|
||
screenspace_points = torch.zeros_like(pc.get_xyz, dtype=pc.get_xyz.dtype, requires_grad=True, device="cuda") + 0
|
||
try:
|
||
# 尝试保留张量的梯度。这是为了确保可以在反向传播过程中计算对于屏幕空间坐标的梯度
|
||
screenspace_points.retain_grad()
|
||
except:
|
||
pass
|
||
|
||
# Set up rasterization configuration
|
||
# 计算视场的 tan 值，这将用于设置光栅化配置
|
||
tanfovx = math.tan(viewpoint_camera.FoVx * 0.5)
|
||
tanfovy = math.tan(viewpoint_camera.FoVy * 0.5)
|
||
|
||
# 设置光栅化的配置，包括图像的大小、视场的 tan 值、背景颜色、视图矩阵viewmatrix、投影矩阵projmatrix等
|
||
raster_settings = GaussianRasterizationSettings(
|
||
image_height=int(viewpoint_camera.image_height),
|
||
image_width=int(viewpoint_camera.image_width),
|
||
tanfovx=tanfovx,
|
||
tanfovy=tanfovy,
|
||
bg=bg_color,
|
||
scale_modifier=scaling_modifier,
|
||
viewmatrix=viewpoint_camera.world_view_transform,
|
||
projmatrix=viewpoint_camera.full_proj_transform,
|
||
sh_degree=pc.active_sh_degree,
|
||
campos=viewpoint_camera.camera_center,
|
||
prefiltered=False,
|
||
debug=pipe.debug
|
||
)
|
||
|
||
# 创建一个高斯光栅化器对象，用于将高斯分布投影到屏幕上
|
||
rasterizer = GaussianRasterizer(raster_settings=raster_settings)
|
||
|
||
# 获取高斯模型的三维坐标、屏幕空间坐标、透明度
|
||
means3D = pc.get_xyz
|
||
means2D = screenspace_points
|
||
opacity = pc.get_opacity
|
||
|
||
# If precomputed 3d covariance is provided, use it. If not, then it will be computed from scaling / rotation by the rasterizer.
|
||
# 如果提供了预先计算的3D协方差矩阵，则使用它。否则，它将由光栅化器根据尺度和旋转进行计算
|
||
scales = None
|
||
rotations = None
|
||
cov3D_precomp = None
|
||
if pipe.compute_cov3D_python:
|
||
cov3D_precomp = pc.get_covariance(scaling_modifier)
|
||
else:
|
||
scales = pc.get_scaling
|
||
rotations = pc.get_rotation
|
||
|
||
# If precomputed colors are provided, use them. Otherwise, if it is desired to precompute colors
|
||
# from SHs in Python, do it. If not, then SH -> RGB conversion will be done by rasterizer.
|
||
# 如果提供了预先计算的颜色，则使用它们。否则，如果希望在Python中从球谐函数中预计算颜色，请执行此操作。如果没有，则颜色将通过光栅化器进行从球谐函数到RGB的转换
|
||
shs = None
|
||
colors_precomp = None
|
||
if override_color is None:
|
||
if pipe.convert_SHs_python:
|
||
# 将SH特征的形状调整为（batch_size * num_points，3，(max_sh_degree+1)**2）
|
||
shs_view = pc.get_features.transpose(1, 2).view(-1, 3, (pc.max_sh_degree+1)**2)
|
||
# 计算相机中心到每个点的方向向量，并归一化
|
||
dir_pp = (pc.get_xyz - viewpoint_camera.camera_center.repeat(pc.get_features.shape[0], 1))
|
||
# 计算相机中心到每个点的方向向量，并归一化
|
||
dir_pp_normalized = dir_pp/dir_pp.norm(dim=1, keepdim=True)
|
||
# 使用SH特征将方向向量转换为RGB颜色
|
||
sh2rgb = eval_sh(pc.active_sh_degree, shs_view, dir_pp_normalized)
|
||
# 将RGB颜色的范围限制在0到1之间
|
||
colors_precomp = torch.clamp_min(sh2rgb + 0.5, 0.0)
|
||
else:
|
||
shs = pc.get_features
|
||
else:
|
||
colors_precomp = override_color
|
||
|
||
# Rasterize visible Gaussians to image, obtain their radii (on screen).
|
||
# 调用光栅化器，将高斯分布投影到屏幕上，获得渲染图像和每个高斯分布在屏幕上的半径
|
||
rendered_image, radii = rasterizer(
|
||
means3D = means3D,
|
||
means2D = means2D,
|
||
shs = shs,
|
||
colors_precomp = colors_precomp,
|
||
opacities = opacity,
|
||
scales = scales,
|
||
rotations = rotations,
|
||
cov3D_precomp = cov3D_precomp)
|
||
|
||
# Those Gaussians that were frustum culled or had a radius of 0 were not visible.
|
||
# They will be excluded from value updates used in the splitting criteria.
|
||
# 返回一个字典，包含渲染的图像、屏幕空间坐标、可见性过滤器（根据半径判断是否可见）以及每个高斯分布在屏幕上的半径
|
||
return_dict={"render": rendered_image,
|
||
"viewspace_points": screenspace_points,
|
||
"visibility_filter" : radii > 0,
|
||
"radii": radii}
|
||
|
||
if return_depth:
|
||
# 提取相机的世界视图变换矩阵的第3列(深度方向)的前3个元素和最后一个元素。这些值将用于计算深度信息
|
||
projvect1 = viewpoint_camera.world_view_transform[:, 2][:3].detach()
|
||
projvect2 = viewpoint_camera.world_view_transform[:, 2][-1].detach()
|
||
# 计算每个3D点的深度值： means3D * projvect1.unsqueeze(0) 将 means3D (3D点坐标) 与 projvect1 (深度方向的前3个元素) 相乘,得到深度分量；
|
||
# 深度分量求和,并加上 projvect2 (深度方向的最后一个元素),得到最终的深度值
|
||
means3D_depth = (means3D * projvect1.unsqueeze(0)).sum(dim=-1, keepdim=True) + projvect2
|
||
# 深度值复制成3个通道,与 colors_precomp 的尺寸匹配
|
||
means3D_depth = means3D_depth.repeat(1, 3)
|
||
render_depth, _ = rasterizer(
|
||
means3D=means3D,
|
||
means2D=means2D,
|
||
shs=None,
|
||
colors_precomp=means3D_depth,
|
||
opacities=opacity,
|
||
scales=scales,
|
||
rotations=rotations,
|
||
cov3D_precomp=cov3D_precomp)
|
||
render_depth = render_depth.mean(dim=0) # 将批量维度上的深度值取平均,得到最终的深度图
|
||
return_dict.update({'render_depth': render_depth})
|
||
|
||
# plt.figure(figsize=(8, 6))
|
||
# plt.imshow(rendered_image.permute(1, 2, 0).detach().cpu().numpy(), cmap='viridis')
|
||
# plt.colorbar()
|
||
# plt.title('Rendered Normal Map')
|
||
# plt.show()
|
||
|
||
# plt.figure(figsize=(8, 6))
|
||
# plt.imshow(render_depth.detach().cpu().numpy(), cmap='viridis')
|
||
# plt.colorbar()
|
||
# plt.title('Rendered Depth Map')
|
||
# plt.show()
|
||
|
||
|
||
if return_normal:
|
||
rotations_mat = build_rotation(rotations)
|
||
scales = pc.get_scaling
|
||
min_scales = torch.argmin(scales, dim=1)
|
||
indices = torch.arange(min_scales.shape[0])
|
||
normal = rotations_mat[indices, :, min_scales]
|
||
|
||
# convert normal direction to the camera; calculate the normal in the camera coordinate
|
||
view_dir = means3D - viewpoint_camera.camera_center
|
||
normal = normal * ((((view_dir * normal).sum(dim=-1) < 0) * 1 - 0.5) * 2)[..., None]
|
||
|
||
R_w2c = torch.tensor(viewpoint_camera.R.T).cuda().to(torch.float32)
|
||
normal = (R_w2c @ normal.transpose(0, 1)).transpose(0, 1)
|
||
|
||
render_normal, _ = rasterizer(
|
||
means3D=means3D,
|
||
means2D=means2D,
|
||
shs=None,
|
||
colors_precomp=normal,
|
||
opacities=opacity,
|
||
scales=scales,
|
||
rotations=rotations,
|
||
cov3D_precomp=cov3D_precomp)
|
||
render_normal = F.normalize(render_normal, dim=0)
|
||
return_dict.update({'render_normal': render_normal})
|
||
|
||
# plt.figure(figsize=(8, 6))
|
||
# plt.imshow(render_normal.permute(1, 2, 0).detach().cpu().numpy(), cmap='viridis')
|
||
# plt.colorbar()
|
||
# plt.title('Rendered Normal Map')
|
||
# plt.show()
|
||
|
||
if return_opacity:
|
||
density = torch.ones_like(means3D)
|
||
|
||
render_opacity, _ = rasterizer(
|
||
means3D=means3D,
|
||
means2D=means2D,
|
||
shs=None,
|
||
colors_precomp=density,
|
||
opacities=opacity,
|
||
scales=scales,
|
||
rotations=rotations,
|
||
cov3D_precomp=cov3D_precomp)
|
||
return_dict.update({'render_opacity': render_opacity.mean(dim=0)})
|
||
|
||
return return_dict |