street_crafter/render.py at main · LiAutoAD/street_crafter · GitHub

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
import torch
import os
import json
import numpy as np
import imageio
from tqdm import tqdm
from street_gaussian.utils.general_utils import safe_state
from street_gaussian.config import cfg
from street_gaussian.visualizers.street_gaussian_visualizer import StreetGaussianisualizer
from street_gaussian.models.street_gaussian_model import StreetGaussianModel
from street_gaussian.models.street_gaussian_renderer import StreetGaussianRenderer
from street_gaussian.utils.diffusion_utils import getDiffusionRunner
from create_scene import create_scene
from easyvolcap.utils.console_utils import *
from easyvolcap.utils.timer_utils import timer

timer.disabled = False

def render_trajectory():
    cfg.render.save_image = False
    cfg.render.save_video = True

    with torch.no_grad():
        scene = create_scene()
        gaussians: StreetGaussianModel = scene.gaussians
        renderer = StreetGaussianRenderer()

        save_dir = os.path.join(cfg.model_path, 'trajectory', "ours_{}".format(scene.loaded_iter))
        visualizer = StreetGaussianisualizer(save_dir)

        train_cameras = scene.getTrainCameras()
        test_cameras = scene.getTestCameras()
        cameras = train_cameras + test_cameras
        cameras = list(sorted(cameras, key=lambda x: x.id))

        for idx, camera in enumerate(tqdm(cameras, desc="Rendering Trajectory")):
            result = renderer.render_all(camera, gaussians)
            visualizer.visualize(result, camera)

        visualizer.summarize()


def render_novel_view():
    cfg.render.save_image = False
    cfg.render.save_video = True

    with torch.no_grad():
        scene = create_scene()
        gaussians: StreetGaussianModel = scene.gaussians
        renderer = StreetGaussianRenderer()

        save_dir = os.path.join(cfg.model_path, 'novel_view', cfg.render.novel_view.name)
        novel_view_cfg = dict(cfg.render.novel_view)
        os.makedirs(save_dir, exist_ok=True)
        with open(os.path.join(save_dir, 'config.json'), 'w') as f:
            json.dump(novel_view_cfg, f, indent=1)

        visualizer = StreetGaussianisualizer(save_dir)

        cameras = scene.getNovelViewCameras()
        assert cameras is not None

        novel_view_ids = list(set([camera.meta['novel_view_id'] for camera in cameras]))
        for novel_view_id in novel_view_ids:
            print(f'Rendering novel view sequence {novel_view_id}')
            cur_cameras = [camera for camera in cameras if camera.meta['novel_view_id'] == novel_view_id]
            cur_cameras = list(sorted(cur_cameras, key=lambda x: x.meta['frame']))
            for idx, camera in enumerate(tqdm(cur_cameras, desc=f"Rendering novel view sequence {novel_view_id}")):
                result = renderer.render_novel_view(camera, gaussians)
                visualizer.visualize_novel_view(result, camera)

            visualizer.result_dir = save_dir + f"/{novel_view_id}"
            os.makedirs(visualizer.result_dir, exist_ok=True)
            visualizer.summarize()
            visualizer.reset()


def run_diffusion():
    cfg.render.save_image = cfg.eval.visualize
    cfg.render.save_video = True

    with torch.no_grad():
        scene = create_scene()
        novel_cameras = scene.getNovelViewCameras()
        train_cameras = scene.getTrainCameras()

        diffusionrunner = getDiffusionRunner(scene)

        # Process novel view sequences
        if not cfg.eval.skip_novel:
            novel_view_ids = list(set([camera.meta['novel_view_id'] for camera in novel_cameras]))

            for novel_view_id in novel_view_ids:
                print(f'Running diffusion for novel view sequence {novel_view_id}')
                cur_cameras = [camera for camera in novel_cameras if camera.meta['novel_view_id'] == novel_view_id]
                cur_cameras = list(sorted(cur_cameras, key=lambda x: x.meta['frame']))

                diffusion_result = diffusionrunner.run(cur_cameras, train_cameras, use_render=False, scale=1.0)

                # Save video
                diffusion_result = (diffusion_result.cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
                diffusion_result = np.transpose(diffusion_result, (0, 2, 3, 1))  # [num_frames, h, 3*w, 3]

                save_dir = os.path.join(cfg.model_path, 'diffusion')
                os.makedirs(save_dir, exist_ok=True)
                save_path = os.path.join(save_dir, f'diffusion_novel_{novel_view_id}.mp4')
                imageio.mimwrite(save_path, diffusion_result, fps=10) # type: ignore

@torch.no_grad()
@catch_throw
def main():
    print("Rendering " + cfg.model_path)
    safe_state(cfg.eval.quiet)

    if cfg.mode == 'trajectory':
        render_trajectory()
    elif cfg.mode == 'novel_view':
        render_novel_view()
    elif cfg.mode == 'diffusion':
        run_diffusion()
    else:
        raise NotImplementedError()


if __name__ == "__main__":
    main()