File size: 6,161 Bytes
917a889
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a4ff4ac
 
 
 
 
5544f1a
 
917a889
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a1e3f5f
 
917a889
 
a1e3f5f
917a889
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import numpy as np
from tqdm import tqdm
import utils3d
from PIL import Image

from ..renderers import MeshRenderer, VoxelRenderer, PbrMeshRenderer
from ..representations import Mesh, Voxel, MeshWithPbrMaterial, MeshWithVoxel
from .random_utils import sphere_hammersley_sequence


def yaw_pitch_r_fov_to_extrinsics_intrinsics(yaws, pitchs, rs, fovs):
    is_list = isinstance(yaws, list)
    if not is_list:
        yaws = [yaws]
        pitchs = [pitchs]
    if not isinstance(rs, list):
        rs = [rs] * len(yaws)
    if not isinstance(fovs, list):
        fovs = [fovs] * len(yaws)
    extrinsics = []
    intrinsics = []
    for yaw, pitch, r, fov in zip(yaws, pitchs, rs, fovs):
        fov = torch.deg2rad(torch.tensor(float(fov))).cuda()
        yaw = torch.tensor(float(yaw)).cuda()
        pitch = torch.tensor(float(pitch)).cuda()
        orig = torch.tensor([
            torch.sin(yaw) * torch.cos(pitch),
            torch.cos(yaw) * torch.cos(pitch),
            torch.sin(pitch),
        ]).cuda() * r
        extr = utils3d.torch.extrinsics_look_at(orig, torch.tensor([0, 0, 0]).float().cuda(), torch.tensor([0, 0, 1]).float().cuda())
        intr = utils3d.torch.intrinsics_from_fov_xy(fov, fov)
        extrinsics.append(extr)
        intrinsics.append(intr)
    if not is_list:
        extrinsics = extrinsics[0]
        intrinsics = intrinsics[0]
    return extrinsics, intrinsics


def get_renderer(sample, **kwargs):
    if isinstance(sample, (MeshWithPbrMaterial, MeshWithVoxel)):
        renderer = PbrMeshRenderer()
        renderer.rendering_options.resolution = kwargs.get('resolution', 512)
        renderer.rendering_options.near = kwargs.get('near', 1)
        renderer.rendering_options.far = kwargs.get('far', 100)
        renderer.rendering_options.ssaa = kwargs.get('ssaa', 2)
        renderer.rendering_options.peel_layers = kwargs.get('peel_layers', 8)
    elif isinstance(sample, Mesh):
        renderer = MeshRenderer()
        renderer.rendering_options.resolution = kwargs.get('resolution', 512)
        renderer.rendering_options.near = kwargs.get('near', 1)
        renderer.rendering_options.far = kwargs.get('far', 100)
        renderer.rendering_options.ssaa = kwargs.get('ssaa', 2)
        renderer.rendering_options.chunk_size = kwargs.get('chunk_size', None)
    elif isinstance(sample, Voxel):
        renderer = VoxelRenderer()
        renderer.rendering_options.resolution = kwargs.get('resolution', 512)
        renderer.rendering_options.near = kwargs.get('near', 0.1)
        renderer.rendering_options.far = kwargs.get('far', 10.0)
        renderer.rendering_options.ssaa = kwargs.get('ssaa', 2)
    else:
        raise ValueError(f'Unsupported sample type: {type(sample)}')
    return renderer


def render_frames(sample, extrinsics, intrinsics, options={}, verbose=True, **kwargs):
    renderer = get_renderer(sample, **options)
    rets = {}
    for j, (extr, intr) in tqdm(enumerate(zip(extrinsics, intrinsics)), total=len(extrinsics), desc='Rendering', disable=not verbose):
        res = renderer.render(sample, extr, intr, **kwargs)
        for k, v in res.items():
            if k not in rets: rets[k] = []
            if v.dim() == 2: v = v[None].repeat(3, 1, 1)
            rets[k].append(np.clip(v.detach().cpu().numpy().transpose(1, 2, 0) * 255, 0, 255).astype(np.uint8))
    return rets


def render_video(sample, resolution=1024, bg_color=(0, 0, 0), num_frames=120, r=2, fov=36, **kwargs):
    # Match render_snapshot angles: start at 150° (5π/6), fixed 20° pitch
    # This gives consistent camera perspective between video and static previews
    start_yaw = 5 * np.pi / 6  # 150 degrees
    pitch_val = 20 / 180 * np.pi  # 20 degrees fixed
    yaws = torch.linspace(start_yaw, 2 * np.pi + start_yaw, num_frames).tolist()
    pitch = [pitch_val] * num_frames
    extrinsics, intrinsics = yaw_pitch_r_fov_to_extrinsics_intrinsics(yaws, pitch, r, fov)
    return render_frames(sample, extrinsics, intrinsics, {'resolution': resolution, 'bg_color': bg_color}, **kwargs)


def render_multiview(sample, resolution=512, nviews=30):
    r = 2
    fov = 40
    cams = [sphere_hammersley_sequence(i, nviews) for i in range(nviews)]
    yaws = [cam[0] for cam in cams]
    pitchs = [cam[1] for cam in cams]
    extrinsics, intrinsics = yaw_pitch_r_fov_to_extrinsics_intrinsics(yaws, pitchs, r, fov)
    res = render_frames(sample, extrinsics, intrinsics, {'resolution': resolution, 'bg_color': (0, 0, 0)})
    return res['color'], extrinsics, intrinsics


def render_snapshot(samples, resolution=512, bg_color=(0, 0, 0), offset=(-16 / 180 * np.pi, 20 / 180 * np.pi), r=10, fov=8, nviews=4, **kwargs):
    yaw = np.linspace(0, 2 * np.pi, nviews, endpoint=False)
    yaw_offset = offset[0]
    yaw = [y + yaw_offset for y in yaw]
    pitch = [offset[1] for _ in range(nviews)]
    extrinsics, intrinsics = yaw_pitch_r_fov_to_extrinsics_intrinsics(yaw, pitch, r, fov)
    return render_frames(samples, extrinsics, intrinsics, {'resolution': resolution, 'bg_color': bg_color}, **kwargs)


def make_pbr_vis_frames(result, resolution=1024):
    num_frames = len(result['shaded'])
    frames = []
    for i in range(num_frames):
        shaded = Image.fromarray(result['shaded'][i])
        normal = Image.fromarray(result['normal'][i])
        base_color = Image.fromarray(result['base_color'][i])
        metallic = Image.fromarray(result['metallic'][i])
        roughness = Image.fromarray(result['roughness'][i])
        alpha = Image.fromarray(result['alpha'][i])
        shaded = shaded.resize((resolution, resolution))
        normal = normal.resize((resolution, resolution))
        base_color = base_color.resize((resolution//2, resolution//2))
        metallic = metallic.resize((resolution//2, resolution//2))
        roughness = roughness.resize((resolution//2, resolution//2))
        alpha = alpha.resize((resolution//2, resolution//2))
        row1 = np.concatenate([shaded, normal], axis=1)
        row2 = np.concatenate([base_color, metallic, roughness, alpha], axis=1)
        frame = np.concatenate([row1, row2], axis=0)
        frames.append(frame)
    return frames