File size: 8,122 Bytes
db6a3b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
#
# 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  [email protected]
#

import torch
import math
from easydict import EasyDict as edict
import numpy as np
from ..representations.gaussian import Gaussian
from .sh_utils import eval_sh
import torch.nn.functional as F
from easydict import EasyDict as edict


def intrinsics_to_projection(
        intrinsics: torch.Tensor,
        near: float,
        far: float,
    ) -> torch.Tensor:
    """
    OpenCV intrinsics to OpenGL perspective matrix

    Args:
        intrinsics (torch.Tensor): [3, 3] OpenCV intrinsics matrix
        near (float): near plane to clip
        far (float): far plane to clip
    Returns:
        (torch.Tensor): [4, 4] OpenGL perspective matrix
    """
    fx, fy = intrinsics[0, 0], intrinsics[1, 1]
    cx, cy = intrinsics[0, 2], intrinsics[1, 2]
    ret = torch.zeros((4, 4), dtype=intrinsics.dtype, device=intrinsics.device)
    ret[0, 0] = 2 * fx
    ret[1, 1] = 2 * fy
    ret[0, 2] = 2 * cx - 1
    ret[1, 2] = - 2 * cy + 1
    ret[2, 2] = far / (far - near)
    ret[2, 3] = near * far / (near - far)
    ret[3, 2] = 1.
    return ret


def render(viewpoint_camera, pc : Gaussian, pipe, bg_color : torch.Tensor, scaling_modifier = 1.0, override_color = None):
    """
    Render the scene. 
    
    Background tensor (bg_color) must be on GPU!
    """
    # lazy import
    if 'GaussianRasterizer' not in globals():
        from diff_gaussian_rasterization import GaussianRasterizer, GaussianRasterizationSettings
    
    # Create zero tensor. We will use it to make pytorch return gradients of the 2D (screen-space) means
    screenspace_points = torch.zeros_like(pc.get_xyz, dtype=pc.get_xyz.dtype, requires_grad=True, device="cuda") + 0
    try:
        screenspace_points.retain_grad()
    except:
        pass
    # Set up rasterization configuration
    tanfovx = math.tan(viewpoint_camera.FoVx * 0.5)
    tanfovy = math.tan(viewpoint_camera.FoVy * 0.5)
    
    kernel_size = pipe.kernel_size
    subpixel_offset = torch.zeros((int(viewpoint_camera.image_height), int(viewpoint_camera.image_width), 2), dtype=torch.float32, device="cuda")

    raster_settings = GaussianRasterizationSettings(
        image_height=int(viewpoint_camera.image_height),
        image_width=int(viewpoint_camera.image_width),
        tanfovx=tanfovx,
        tanfovy=tanfovy,
        kernel_size=kernel_size,
        subpixel_offset=subpixel_offset,
        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
    )
    
    rasterizer = GaussianRasterizer(raster_settings=raster_settings)

    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.
    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.
    shs = None
    colors_precomp = None
    if override_color is None:
        if pipe.convert_SHs_python:
            shs_view = pc.get_features.transpose(1, 2).view(-1, 3, (pc.max_sh_degree+1)**2)
            dir_pp = (pc.get_xyz - viewpoint_camera.camera_center.repeat(pc.get_features.shape[0], 1))
            dir_pp_normalized = dir_pp/dir_pp.norm(dim=1, keepdim=True)
            sh2rgb = eval_sh(pc.active_sh_degree, shs_view, dir_pp_normalized)
            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). 
    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.
    return edict({"render": rendered_image,
            "viewspace_points": screenspace_points,
            "visibility_filter" : radii > 0,
            "radii": radii})


class GaussianRenderer:
    """
    Renderer for the Voxel representation.

    Args:
        rendering_options (dict): Rendering options.
    """

    def __init__(self, rendering_options={}) -> None:
        self.pipe = edict({
            "kernel_size": 0.1,
            "convert_SHs_python": False,
            "compute_cov3D_python": False,
            "scale_modifier": 1.0,
            "debug": False
        })
        self.rendering_options = edict({
            "resolution": None,
            "near": None,
            "far": None,
            "ssaa": 1,
            "bg_color": 'random',
        })
        self.rendering_options.update(rendering_options)
        self.bg_color = None
    
    def render(
            self,
            gausssian: Gaussian,
            extrinsics: torch.Tensor,
            intrinsics: torch.Tensor,
            colors_overwrite: torch.Tensor = None
        ) -> edict:
        """
        Render the gausssian.

        Args:
            gaussian : gaussianmodule
            extrinsics (torch.Tensor): (4, 4) camera extrinsics
            intrinsics (torch.Tensor): (3, 3) camera intrinsics
            colors_overwrite (torch.Tensor): (N, 3) override color

        Returns:
            edict containing:
                color (torch.Tensor): (3, H, W) rendered color image
        """
        resolution = self.rendering_options["resolution"]
        near = self.rendering_options["near"]
        far = self.rendering_options["far"]
        ssaa = self.rendering_options["ssaa"]
        
        if self.rendering_options["bg_color"] == 'random':
            self.bg_color = torch.zeros(3, dtype=torch.float32, device="cuda")
            if np.random.rand() < 0.5:
                self.bg_color += 1
        else:
            self.bg_color = torch.tensor(self.rendering_options["bg_color"], dtype=torch.float32, device="cuda")

        view = extrinsics
        perspective = intrinsics_to_projection(intrinsics, near, far)
        camera = torch.inverse(view)[:3, 3]
        focalx = intrinsics[0, 0]
        focaly = intrinsics[1, 1]
        fovx = 2 * torch.atan(0.5 / focalx)
        fovy = 2 * torch.atan(0.5 / focaly)
            
        camera_dict = edict({
            "image_height": resolution * ssaa,
            "image_width": resolution * ssaa,
            "FoVx": fovx,
            "FoVy": fovy,
            "znear": near,
            "zfar": far,
            "world_view_transform": view.T.contiguous(),
            "projection_matrix": perspective.T.contiguous(),
            "full_proj_transform": (perspective @ view).T.contiguous(),
            "camera_center": camera
        })

        # Render
        render_ret = render(camera_dict, gausssian, self.pipe, self.bg_color, override_color=colors_overwrite, scaling_modifier=self.pipe.scale_modifier)

        if ssaa > 1:
            render_ret.render = F.interpolate(render_ret.render[None], size=(resolution, resolution), mode='bilinear', align_corners=False, antialias=True).squeeze()
            
        ret = edict({
            'color': render_ret['render']
        })
        return ret