learned-point-cloud-compression-for-classification/scripts/convert_dataset.py at main · multimedialabsfu/learned-point-cloud-compression-for-classification · 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
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
import argparse
import subprocess
from pathlib import Path

import numpy as np
from pyntcloud import PyntCloud


def ctmconv(input_file, output_file):
    subprocess.run(["ctmconv", input_file, output_file])


def process_points(args, points):
    assert points.ndim == 2 and points.shape[-1] == 3

    if args.normalize:
        points = points - points.mean(axis=0)
        point_norms = np.sqrt((points**2).sum(axis=1))
        points = points / point_norms.max()

    if args.translate != [0, 0, 0]:
        points = points + np.array(args.translate)

    if args.scale != 1.0:
        points = points * args.scale

    if args.quantize:
        points = points.round()

    return points


def process_cloud(args, cloud):
    xyz = ["x", "y", "z"]
    assert cloud.points.columns.tolist() == xyz

    mesh = cloud.mesh
    cloud.points = cloud.points.astype(np.float64, copy=False)
    cloud.mesh = mesh  # Restore mesh.

    if args.num_points_resample > 0:
        cloud = cloud.get_sample(
            "mesh_random", n=args.num_points_resample, as_PyntCloud=True
        )

    cloud.points[xyz] = process_points(args, cloud.points[xyz].values)
    if args.unique:
        _, idx = np.unique(cloud.points[xyz].values, axis=0, return_index=True)
        cloud.points = cloud.points.iloc[idx]
    if "mesh" in args.output_attributes:
        cloud.mesh = mesh  # Restore mesh.

    if args.num_points_subsample > 0:
        assert len(cloud.points) >= args.num_points_subsample
        idx = np.random.choice(
            len(cloud.points), args.num_points_subsample, replace=False
        )
        cloud.points = cloud.points.iloc[idx]

    return cloud


def iterate_paths(args):
    if args.input_file:
        yield Path(args.input_file), Path(args.output_file)
        return

    input_root = Path(args.input_dir)
    output_root = Path(args.output_dir)

    for input_filepath in input_root.rglob(f"*.{args.input_extension}"):
        filepath = Path(input_filepath).relative_to(input_root)
        output_filepath = (output_root / filepath).with_suffix(
            f".{args.output_extension}"
        )
        output_filepath.parent.mkdir(parents=True, exist_ok=True)
        yield input_filepath, output_filepath


def build_parser():
    parser = argparse.ArgumentParser()
    parser.add_argument("--input_file", type=str)
    parser.add_argument("--input_dir", type=str)
    parser.add_argument("--output_file", type=str)
    parser.add_argument("--output_dir", type=str)
    parser.add_argument("--input_extension", type=str, default="*")
    parser.add_argument("--output_extension", type=str, default="ply")
    parser.add_argument("--convert_only", type=bool, default=False)
    parser.add_argument(
        "--output_attributes",
        type=str,
        help="Attributes to copy from input to output, e.g. 'mesh'. Comma separated.",
        default="",
    )
    parser.add_argument("--normalize", type=bool, default=False)
    parser.add_argument("--translate", type=float, nargs=3, default=[0, 0, 0])
    parser.add_argument("--scale", type=float, default=1.0)
    parser.add_argument("--quantize", type=bool, default=False)
    parser.add_argument("--unique", type=bool, default=False)
    parser.add_argument("--num_points_resample", type=int, default=0)
    parser.add_argument("--num_points_subsample", type=int, default=0)
    return parser


def parse_args(parser, argv=None):
    args = parser.parse_args(argv)

    if bool(args.input_file) == bool(args.input_dir):
        raise ValueError("Please specify exactly one of --input_file or --input_dir.")

    if bool(args.output_file) == bool(args.output_dir):
        raise ValueError("Please specify exactly one of --output_file or --output_dir.")

    if (bool(args.input_file) != bool(args.output_file)) or (
        bool(args.input_dir) != bool(args.output_dir)
    ):
        raise ValueError(
            "Please specify only files or only directories for input and output."
        )

    args.output_attributes = (
        args.output_attributes.strip().split(",") if args.output_attributes else []
    )

    if "mesh" in args.output_attributes:
        raise ValueError("mesh attribute is not supported.")
        if args.num_points_resample > 0:
            raise ValueError("--num_points_resample destroys the mesh attribute.")
        if args.num_points_subsample > 0:
            raise ValueError("--num_points_subsample destroys the mesh attribute.")

    return args


def main(argv=None):
    parser = build_parser()
    args = parse_args(parser, argv)

    for input_filepath, output_filepath in sorted(iterate_paths(args)):
        print(f"Processing\n  in:  {input_filepath}\n  out: {output_filepath}")

        if args.convert_only:
            assert (
                args.output_attributes == []
                and args.normalize is False
                and args.translate == [0, 0, 0]
                and args.scale == 1.0
                and args.quantize is False
                and args.unique is False
                and args.num_points_resample == 0
                and args.num_points_subsample == 0
            )
            ctmconv(input_filepath, output_filepath)
            continue

        cloud = PyntCloud.from_file(str(input_filepath))
        cloud = process_cloud(args, cloud)
        cloud.to_file(str(output_filepath), also_save=args.output_attributes)


if __name__ == "__main__":
    main()