File size: 14,422 Bytes
f53b39e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
# Copyright (C) 2024-present Naver Corporation. All rights reserved.
# Licensed under CC BY-NC-SA 4.0 (non-commercial use only).
#
# --------------------------------------------------------
# gradio demo
# --------------------------------------------------------
import argparse
import math
import builtins
import datetime
import gradio
import os
import torch
import numpy as np
import functools
import trimesh
import copy
from scipy.spatial.transform import Rotation

from dust3r.inference import inference
from dust3r.image_pairs import make_pairs
from dust3r.utils.image import load_images, rgb
from dust3r.utils.device import to_numpy
from dust3r.viz import add_scene_cam, CAM_COLORS, OPENGL, pts3d_to_trimesh, cat_meshes
from dust3r.cloud_opt import global_aligner, GlobalAlignerMode

import matplotlib.pyplot as pl


def get_args_parser():
    parser = argparse.ArgumentParser()
    parser_url = parser.add_mutually_exclusive_group()
    parser_url.add_argument("--local_network", action='store_true', default=False,
                            help="make app accessible on local network: address will be set to 0.0.0.0")
    parser_url.add_argument("--server_name", type=str, default=None, help="server url, default is 127.0.0.1")
    parser.add_argument("--image_size", type=int, default=512, choices=[512, 224], help="image size")
    parser.add_argument("--server_port", type=int, help=("will start gradio app on this port (if available). "
                                                         "If None, will search for an available port starting at 7860."),
                        default=None)
    parser_weights = parser.add_mutually_exclusive_group(required=True)
    parser_weights.add_argument("--weights", type=str, help="path to the model weights", default=None)
    parser_weights.add_argument("--model_name", type=str, help="name of the model weights",
                                choices=["DUSt3R_ViTLarge_BaseDecoder_512_dpt",
                                         "DUSt3R_ViTLarge_BaseDecoder_512_linear",
                                         "DUSt3R_ViTLarge_BaseDecoder_224_linear"])
    parser.add_argument("--device", type=str, default='cuda', help="pytorch device")
    parser.add_argument("--tmp_dir", type=str, default=None, help="value for tempfile.tempdir")
    parser.add_argument("--silent", action='store_true', default=False,
                        help="silence logs")
    return parser


def set_print_with_timestamp(time_format="%Y-%m-%d %H:%M:%S"):
    builtin_print = builtins.print

    def print_with_timestamp(*args, **kwargs):
        now = datetime.datetime.now()
        formatted_date_time = now.strftime(time_format)

        builtin_print(f'[{formatted_date_time}] ', end='')  # print with time stamp
        builtin_print(*args, **kwargs)

    builtins.print = print_with_timestamp


def _convert_scene_output_to_glb(outdir, imgs, pts3d, mask, focals, cams2world, cam_size=0.05,
                                 cam_color=None, as_pointcloud=False,
                                 transparent_cams=False, silent=False):
    assert len(pts3d) == len(mask) <= len(imgs) <= len(cams2world) == len(focals)
    pts3d = to_numpy(pts3d)
    imgs = to_numpy(imgs)
    focals = to_numpy(focals)
    cams2world = to_numpy(cams2world)

    scene = trimesh.Scene()

    # full pointcloud
    if as_pointcloud:
        pts = np.concatenate([p[m] for p, m in zip(pts3d, mask)])
        col = np.concatenate([p[m] for p, m in zip(imgs, mask)])
        pct = trimesh.PointCloud(pts.reshape(-1, 3), colors=col.reshape(-1, 3))
        scene.add_geometry(pct)
    else:
        meshes = []
        for i in range(len(imgs)):
            meshes.append(pts3d_to_trimesh(imgs[i], pts3d[i], mask[i]))
        mesh = trimesh.Trimesh(**cat_meshes(meshes))
        scene.add_geometry(mesh)

    # add each camera
    for i, pose_c2w in enumerate(cams2world):
        if isinstance(cam_color, list):
            camera_edge_color = cam_color[i]
        else:
            camera_edge_color = cam_color or CAM_COLORS[i % len(CAM_COLORS)]
        add_scene_cam(scene, pose_c2w, camera_edge_color,
                      None if transparent_cams else imgs[i], focals[i],
                      imsize=imgs[i].shape[1::-1], screen_width=cam_size)

    rot = np.eye(4)
    rot[:3, :3] = Rotation.from_euler('y', np.deg2rad(180)).as_matrix()
    scene.apply_transform(np.linalg.inv(cams2world[0] @ OPENGL @ rot))
    outfile = os.path.join(outdir, 'scene.glb')
    if not silent:
        print('(exporting 3D scene to', outfile, ')')
    scene.export(file_obj=outfile)
    return outfile


def get_3D_model_from_scene(outdir, silent, scene, min_conf_thr=3, as_pointcloud=False, mask_sky=False,
                            clean_depth=False, transparent_cams=False, cam_size=0.05):
    """
    extract 3D_model (glb file) from a reconstructed scene
    """
    if scene is None:
        return None
    # post processes
    if clean_depth:
        scene = scene.clean_pointcloud()
    if mask_sky:
        scene = scene.mask_sky()

    # get optimized values from scene
    rgbimg = scene.imgs
    focals = scene.get_focals().cpu()
    cams2world = scene.get_im_poses().cpu()
    # 3D pointcloud from depthmap, poses and intrinsics
    pts3d = to_numpy(scene.get_pts3d())
    scene.min_conf_thr = float(scene.conf_trf(torch.tensor(min_conf_thr)))
    msk = to_numpy(scene.get_masks())
    return _convert_scene_output_to_glb(outdir, rgbimg, pts3d, msk, focals, cams2world, as_pointcloud=as_pointcloud,
                                        transparent_cams=transparent_cams, cam_size=cam_size, silent=silent)


def get_reconstructed_scene(outdir, model, device, silent, image_size, filelist, schedule, niter, min_conf_thr,
                            as_pointcloud, mask_sky, clean_depth, transparent_cams, cam_size,
                            scenegraph_type, winsize, refid):
    """
    from a list of images, run dust3r inference, global aligner.
    then run get_3D_model_from_scene
    """
    imgs = load_images(filelist, size=image_size, verbose=not silent)
    if len(imgs) == 1:
        imgs = [imgs[0], copy.deepcopy(imgs[0])]
        imgs[1]['idx'] = 1
    if scenegraph_type == "swin":
        scenegraph_type = scenegraph_type + "-" + str(winsize)
    elif scenegraph_type == "oneref":
        scenegraph_type = scenegraph_type + "-" + str(refid)

    pairs = make_pairs(imgs, scene_graph=scenegraph_type, prefilter=None, symmetrize=True)
    output = inference(pairs, model, device, batch_size=1, verbose=not silent)

    mode = GlobalAlignerMode.PointCloudOptimizer if len(imgs) > 2 else GlobalAlignerMode.PairViewer
    scene = global_aligner(output, device=device, mode=mode, verbose=not silent)
    lr = 0.01

    if mode == GlobalAlignerMode.PointCloudOptimizer:
        loss = scene.compute_global_alignment(init='mst', niter=niter, schedule=schedule, lr=lr)

    outfile = get_3D_model_from_scene(outdir, silent, scene, min_conf_thr, as_pointcloud, mask_sky,
                                      clean_depth, transparent_cams, cam_size)

    # also return rgb, depth and confidence imgs
    # depth is normalized with the max value for all images
    # we apply the jet colormap on the confidence maps
    rgbimg = scene.imgs
    depths = to_numpy(scene.get_depthmaps())
    confs = to_numpy([c for c in scene.im_conf])
    cmap = pl.get_cmap('jet')
    depths_max = max([d.max() for d in depths])
    depths = [d / depths_max for d in depths]
    confs_max = max([d.max() for d in confs])
    confs = [cmap(d / confs_max) for d in confs]

    imgs = []
    for i in range(len(rgbimg)):
        imgs.append(rgbimg[i])
        imgs.append(rgb(depths[i]))
        imgs.append(rgb(confs[i]))

    return scene, outfile, imgs


def set_scenegraph_options(inputfiles, winsize, refid, scenegraph_type):
    num_files = len(inputfiles) if inputfiles is not None else 1
    max_winsize = max(1, math.ceil((num_files - 1) / 2))
    if scenegraph_type == "swin":
        winsize = gradio.Slider(label="Scene Graph: Window Size", value=max_winsize,
                                minimum=1, maximum=max_winsize, step=1, visible=True)
        refid = gradio.Slider(label="Scene Graph: Id", value=0, minimum=0,
                              maximum=num_files - 1, step=1, visible=False)
    elif scenegraph_type == "oneref":
        winsize = gradio.Slider(label="Scene Graph: Window Size", value=max_winsize,
                                minimum=1, maximum=max_winsize, step=1, visible=False)
        refid = gradio.Slider(label="Scene Graph: Id", value=0, minimum=0,
                              maximum=num_files - 1, step=1, visible=True)
    else:
        winsize = gradio.Slider(label="Scene Graph: Window Size", value=max_winsize,
                                minimum=1, maximum=max_winsize, step=1, visible=False)
        refid = gradio.Slider(label="Scene Graph: Id", value=0, minimum=0,
                              maximum=num_files - 1, step=1, visible=False)
    return winsize, refid


def main_demo(tmpdirname, model, device, image_size, server_name, server_port, silent=False):
    recon_fun = functools.partial(get_reconstructed_scene, tmpdirname, model, device, silent, image_size)
    model_from_scene_fun = functools.partial(get_3D_model_from_scene, tmpdirname, silent)
    with gradio.Blocks(css=""".gradio-container {margin: 0 !important; min-width: 100%};""", title="DUSt3R Demo") as demo:
        # scene state is save so that you can change conf_thr, cam_size... without rerunning the inference
        scene = gradio.State(None)
        gradio.HTML('<h2 style="text-align: center;">DUSt3R Demo</h2>')
        with gradio.Column():
            inputfiles = gradio.File(file_count="multiple")
            with gradio.Row():
                schedule = gradio.Dropdown(["linear", "cosine"],
                                           value='linear', label="schedule", info="For global alignment!")
                niter = gradio.Number(value=300, precision=0, minimum=0, maximum=5000,
                                      label="num_iterations", info="For global alignment!")
                scenegraph_type = gradio.Dropdown([("complete: all possible image pairs", "complete"),
                                                   ("swin: sliding window", "swin"),
                                                   ("oneref: match one image with all", "oneref")],
                                                  value='complete', label="Scenegraph",
                                                  info="Define how to make pairs",
                                                  interactive=True)
                winsize = gradio.Slider(label="Scene Graph: Window Size", value=1,
                                        minimum=1, maximum=1, step=1, visible=False)
                refid = gradio.Slider(label="Scene Graph: Id", value=0, minimum=0, maximum=0, step=1, visible=False)

            run_btn = gradio.Button("Run")

            with gradio.Row():
                # adjust the confidence threshold
                min_conf_thr = gradio.Slider(label="min_conf_thr", value=3.0, minimum=1.0, maximum=20, step=0.1)
                # adjust the camera size in the output pointcloud
                cam_size = gradio.Slider(label="cam_size", value=0.05, minimum=0.001, maximum=0.1, step=0.001)
            with gradio.Row():
                as_pointcloud = gradio.Checkbox(value=False, label="As pointcloud")
                # two post process implemented
                mask_sky = gradio.Checkbox(value=False, label="Mask sky")
                clean_depth = gradio.Checkbox(value=True, label="Clean-up depthmaps")
                transparent_cams = gradio.Checkbox(value=False, label="Transparent cameras")

            outmodel = gradio.Model3D()
            outgallery = gradio.Gallery(label='rgb,depth,confidence', columns=3, height="100%")

            # events
            scenegraph_type.change(set_scenegraph_options,
                                   inputs=[inputfiles, winsize, refid, scenegraph_type],
                                   outputs=[winsize, refid])
            inputfiles.change(set_scenegraph_options,
                              inputs=[inputfiles, winsize, refid, scenegraph_type],
                              outputs=[winsize, refid])
            run_btn.click(fn=recon_fun,
                          inputs=[inputfiles, schedule, niter, min_conf_thr, as_pointcloud,
                                  mask_sky, clean_depth, transparent_cams, cam_size,
                                  scenegraph_type, winsize, refid],
                          outputs=[scene, outmodel, outgallery])
            min_conf_thr.release(fn=model_from_scene_fun,
                                 inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                         clean_depth, transparent_cams, cam_size],
                                 outputs=outmodel)
            cam_size.change(fn=model_from_scene_fun,
                            inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                    clean_depth, transparent_cams, cam_size],
                            outputs=outmodel)
            as_pointcloud.change(fn=model_from_scene_fun,
                                 inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                         clean_depth, transparent_cams, cam_size],
                                 outputs=outmodel)
            mask_sky.change(fn=model_from_scene_fun,
                            inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                    clean_depth, transparent_cams, cam_size],
                            outputs=outmodel)
            clean_depth.change(fn=model_from_scene_fun,
                               inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                       clean_depth, transparent_cams, cam_size],
                               outputs=outmodel)
            transparent_cams.change(model_from_scene_fun,
                                    inputs=[scene, min_conf_thr, as_pointcloud, mask_sky,
                                            clean_depth, transparent_cams, cam_size],
                                    outputs=outmodel)
    demo.launch(share=False, server_name=server_name, server_port=server_port)