self-supervised-depth-denoising/src/data_processing/augment_in_3d.py at master · alr-internship/self-supervised-depth-denoising · 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

import copy
from pathlib import Path
from typing import Tuple
from joblib import Parallel, delayed
import numpy as np
import open3d as o3d
from matplotlib import pyplot as plt
from argparse import ArgumentParser
from psutil import cpu_count

from tqdm import tqdm
from dataset.dataset_interface import DatasetInterface
from utils.general_utils import split
from utils.transformation_utils import imgs_to_pcd, pcd_to_imgs, rs_ci, fill_to_shape, to_rgb


def sample_transformation():
    # rotation settings
    rotate_low = -5 # in degree
    rotate_high = 5 # in degree
    rotate = np.random.rand(3) * (rotate_high - rotate_low) + rotate_low
    rotate = np.deg2rad(rotate)
    R = o3d.geometry.get_rotation_matrix_from_xyz(rotate)

    # translation
    translate_min = [-0.05, -0.05, -0.05] # in meter
    translate_max = [0.05, 0.05, 0.05]
    translation = [np.random.rand(1)\
        * (t_max - t_min) + t_min
        for t_min, t_max in zip (translate_min, translate_max)
    ]


    def transform(pcds):
        transformed_pcds = []
        for pcd in pcds:
            pcd = copy.deepcopy(pcd)
            pcd.rotate(R, center=pcds[0].get_center())
            pcd.translate(translation)
            transformed_pcds.append(pcd)
        return transformed_pcds

    return transform


def augment_files(files, in_dir, out_dir, num_augs_per_img):
    for file in tqdm(files):
        imgs = DatasetInterface.load(file)
        out_dir_path = out_dir / file.relative_to(in_dir).parent

        # augmented_img_sets = generate_augmentations(imgs, num_augs=num_augs_per_img)
        rs_rgb, rs_depth, zv_rgb, zv_depth, mask = imgs

        if len(mask.shape) == 2:
            mask = mask[..., None]

        mask = np.sum(mask, axis=2, keepdims=True) > 0

        rgb_mask = np.repeat(mask.astype(np.uint8), repeats=3, axis=2)

        rs_pcd = imgs_to_pcd(rs_rgb, rs_depth.astype(np.float32), rs_ci)
        zv_pcd = imgs_to_pcd(zv_rgb, zv_depth.astype(np.float32), rs_ci)
        mask_pcd = imgs_to_pcd(rgb_mask, zv_depth.astype(np.float32), rs_ci)
        # o3d.visualization.draw_geometries([rs_pcd, zv_pcd])

        # _, ax = plt.subplots(3, num_augs + 1)

        # ax[0][0].imshow(rs_rgb)
        # ax[1][0].imshow(zv_rgb)
        # ax[2][0].imshow(mask)
        orignal_imgs = (rs_rgb, rs_depth, zv_rgb, zv_depth, mask)
        original_out_path = out_dir_path / file.name
        if not original_out_path.exists():
            DatasetInterface.save(*orignal_imgs, file_name=original_out_path)

        for idx in range(num_augs_per_img):
            out_file_name = out_dir_path / f"{file.stem}_{idx}.npz"
            if out_file_name.exists():
                continue

            transform = sample_transformation()
            t_rs_pcd, t_zv_pcd, t_mask_pcd = transform([rs_pcd, zv_pcd, mask_pcd])
            # o3d.visualization.draw_geometries([t_rs_pcd, t_zv_pcd])

            t_rs_rgb, t_rs_depth, _, _ = pcd_to_imgs(t_rs_pcd, rs_ci)
            t_zv_rgb, t_zv_depth, _, _ = pcd_to_imgs(t_zv_pcd, rs_ci)
            t_mask, _, _, _ = pcd_to_imgs(t_mask_pcd, rs_ci)

            t_mask = t_mask[..., 0]
            t_mask = t_mask[..., None]

            t_rs_rgb = fill_to_shape(t_rs_rgb, rs_rgb.shape, 0, dtype=np.uint8)
            t_rs_depth = fill_to_shape(t_rs_depth, rs_depth.shape, np.nan, dtype=np.float32)
            t_zv_rgb = fill_to_shape(t_zv_rgb, zv_rgb.shape, 0, dtype=np.uint8)
            t_zv_depth = fill_to_shape(t_zv_depth, zv_depth.shape, np.nan, dtype=np.float32)
            t_mask = fill_to_shape(t_mask, mask.shape, False, dtype=bool)
            augmented_imgs = (t_rs_rgb, t_rs_depth, t_zv_rgb, t_zv_depth, t_mask)

            DatasetInterface.save(*augmented_imgs, file_name=out_file_name)

            # ax[0][i].imshow(to_rgb(t_rs_rgb))
            # ax[1][i].imshow(to_rgb(t_zv_rgb))
            # ax[2][i].imshow(t_mask)

        # plt.show()


def main(args):
    img_dir = args.in_dir
    out_dir = args.out_dir
    num_augs_per_img = args.num_augs
    jobs = args.jobs

    files = list(img_dir.rglob("*.npz"))

    if jobs > 1:
        files_chunked = split(files, jobs)

        Parallel(jobs)(
            delayed(augment_files)(files_chunk, img_dir, out_dir, num_augs_per_img)
            for files_chunk in files_chunked
        )

    else:
        augment_files(files)


if __name__ == "__main__":
    argparse = ArgumentParser()
    argparse.add_argument("in_dir", type=Path)
    argparse.add_argument("out_dir", type=Path)
    argparse.add_argument("--num-augs", type=int, default=5)
    argparse.add_argument("--jobs", type=int, default=cpu_count())
    main(argparse.parse_args())