File size: 5,922 Bytes
d0e86f6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

import copy
import logging
from typing import Optional

import torch

logger = logging.getLogger(__name__)
import torch.nn.functional as Fu
from pytorch3d.ops import interpolate_face_attributes
from pytorch3d.renderer import (
    AmbientLights,
    BlendParams,
    FoVPerspectiveCameras,
    HardPhongShader,
    Materials,
    MeshRasterizer,
    MeshRendererWithFragments,
    PerspectiveCameras,
    RasterizationSettings,
)
from pytorch3d.structures import Meshes
from torch import nn
from torch.nn import functional as F

EPS = 1e-8
MAX_N_FACES_OPENGL = 25_000_000

try:
    from pytorch3d.renderer.opengl import MeshRasterizerOpenGL

    _OPENGL_BACKEND = True
except ImportError:
    logger.warning("opengl backend not found")
    _OPENGL_BACKEND = False


class Renderer(nn.Module):
    def __init__(self, img_size: int, **kwargs):
        super().__init__()

        self.img_size = (img_size, img_size) if isinstance(img_size, int) else img_size
        self.opengl_backend = kwargs.pop("opengl_backend", False)

        self.lights = AmbientLights()

        r_kwargs = {
            "perspective_correct": kwargs.pop("perspective_correct", None),
            "clip_barycentric_coords": True,
            "z_clip_value": kwargs.pop("z_clip", None),
        }
        raster_settings = RasterizationSettings(
            image_size=(self.img_size[0] * 2, self.img_size[1] * 2),
            bin_size=0,  # rendering bugs for large number of faces with default
            # see https://github.com/facebookresearch/pytorch3d/issues/348
            max_faces_opengl=MAX_N_FACES_OPENGL,
            **r_kwargs,
        )

        if self.opengl_backend and _OPENGL_BACKEND:
            rasterizer_cls = MeshRasterizerOpenGL
        else:
            rasterizer_cls = MeshRasterizer

        s_kwargs = {"cameras": None, "lights": self.lights}
        s_kwargs["blend_params"] = BlendParams(
            background_color=kwargs.pop("background_color", (0, 0, 0)),
            sigma=0,
        )
        shader_cls = HardPhongShaderPlus

        self.renderer = VizMeshRendererWithFragments(
            rasterizer_cls(cameras=None, raster_settings=raster_settings),
            shader_cls(**s_kwargs),
        )

    def to(self, device):
        super().to(device)
        self.renderer = self.renderer.to(device)
        return self

    def forward(
        self,
        meshes: Meshes,
        cameras: PerspectiveCameras,
        return_normals: bool = False,
    ) -> tuple[
        torch.Tensor,
        torch.Tensor,
        Optional[torch.Tensor],
    ]:
        # XXX bug when using perspective_correct, see https://github.com/facebookresearch/pytorch3d/issues/561
        # setting eps fixes the issue
        images, fragments = self.renderer(meshes, eps=EPS, cameras=cameras)
        images, masks = images.split([3, 1], dim=1)

        normal = None
        if return_normals:
            normal = soft_normal_shading(
                fragments=fragments,
                meshes=meshes,
                cameras=cameras,
            )

        # -- Post-processing
        masks = masks.clamp(0.0, 1.0)
        if return_normals:
            normal = normal.clamp(0.0, 1.0)
            normal = make_normal_image(masks, normal)

        return images, masks, normal  # BCHW


class VizMeshRendererWithFragments(MeshRendererWithFragments):
    """Renderer for visualization, with anti-aliasing"""

    @torch.no_grad()
    def __call__(self, *input, **kwargs):
        images, fragments = super().__call__(*input, **kwargs)
        return F.avg_pool2d(images, kernel_size=2, stride=2), fragments


class HardPhongShaderPlus(HardPhongShader):
    """Rewriting to permute output tensor + working `to` method for multi-gpus"""

    def forward(self, fragments, meshes, **kwargs):
        return super().forward(fragments, meshes, **kwargs).permute(0, 3, 1, 2)

    def to(self, device):
        cameras = self.cameras
        if cameras is not None:
            self.cameras = cameras.to(device)
        self.materials = self.materials.to(device)
        self.lights = self.lights.to(device)
        return self


def soft_normal_shading(
    fragments,
    meshes: Meshes,
    cameras: PerspectiveCameras,
    rescale: bool = True,
    **kwargs,
):
    """Soft normal shading using interpolated vertex normals for smooth appearance."""
    cameras_sphere = copy.copy(cameras)
    cameras_sphere.T = cameras_sphere.T / 2.0

    # Use interpolated vertex normals instead of face normals
    faces = meshes.faces_packed()  # (F, 3)
    vertex_normals = meshes.verts_normals_packed()  # (V, 3)
    faces_normals = vertex_normals[faces]  # (F, 3, 3)
    # Interpolate normals using barycentric coordinates
    pixel_normals = interpolate_face_attributes(
        fragments.pix_to_face, fragments.bary_coords, faces_normals
    )
    shape = pixel_normals.shape
    pixel_normals = cameras_sphere.get_world_to_view_transform(
        **kwargs
    ).transform_points(pixel_normals.view(shape[0], -1, 3), eps=kwargs.get("eps", 1e-7))
    pixel_normals = pixel_normals.view(*shape)
    if rescale:
        pixel_normals = (F.normalize(pixel_normals, dim=-1) + 1) * 0.5
    # (N, H, W, K, 3) -> (N, 3, H, W) - take first face per pixel
    pixel_normals = pixel_normals[..., 0, :].permute((0, 3, 1, 2))
    return pixel_normals


def make_normal_image(mask_render: torch.Tensor, normal_render: torch.Tensor):

    mask_cpu = mask_render.cpu().detach().clone()
    normal_cpu = normal_render.cpu().detach().clone()

    normal_resize = Fu.interpolate(
        normal_cpu,
        size=mask_cpu.shape[2:],
        mode="nearest",
    )
    normal_cpu = normal_resize * mask_cpu + (1 - mask_cpu)
    return normal_cpu