File size: 4,855 Bytes
1ea89dd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import json
import os

import h5py
import numpy as np
import torch

from unik3d.datasets.image_dataset import ImageDataset
from unik3d.datasets.sequence_dataset import SequenceDataset
from unik3d.datasets.utils import DatasetFromList


class ETH3D(ImageDataset):
    min_depth = 0.01
    max_depth = 50.0
    depth_scale = 1000.0
    test_split = "train.txt"
    train_split = "train.txt"
    intrisics_file = "intrinsics.json"
    hdf5_paths = ["ETH3D.hdf5"]

    def __init__(
        self,
        image_shape,
        split_file,
        test_mode,
        benchmark=False,
        augmentations_db={},
        normalize=True,
        resize_method="hard",
        mini=1.0,
        **kwargs,
    ):
        super().__init__(
            image_shape=image_shape,
            split_file=split_file,
            test_mode=test_mode,
            benchmark=benchmark,
            normalize=normalize,
            augmentations_db=augmentations_db,
            resize_method=resize_method,
            mini=mini,
            **kwargs,
        )
        self.test_mode = test_mode
        self.load_dataset()

    def load_dataset(self):
        h5file = h5py.File(
            os.path.join(self.data_root, self.hdf5_paths[0]),
            "r",
            libver="latest",
            swmr=True,
        )
        txt_file = np.array(h5file[self.split_file])
        txt_string = txt_file.tostring().decode("ascii")[:-1]  # correct the -1
        intrinsics = np.array(h5file[self.intrisics_file]).tostring().decode("ascii")
        intrinsics = json.loads(intrinsics)
        h5file.close()
        dataset = []
        for line in txt_string.split("\n"):
            image_filename, depth_filename = line.strip().split(" ")
            intrinsics_val = torch.tensor(intrinsics[image_filename]).squeeze()[:, :3]
            sample = [image_filename, depth_filename, intrinsics_val]
            dataset.append(sample)

        self.dataset = DatasetFromList(dataset)
        self.log_load_dataset()

    def pre_pipeline(self, results):
        results = super().pre_pipeline(results)
        results["dense"] = [True] * self.num_frames * self.num_copies
        results["quality"] = [1] * self.num_frames * self.num_copies
        return results


class ETH3D_F(SequenceDataset):
    min_depth = 0.05
    max_depth = 60.0
    depth_scale = 1000.0
    test_split = "train.txt"
    train_split = "train.txt"
    sequences_file = "sequences.json"
    hdf5_paths = ["ETH3D-F.hdf5"]

    def __init__(
        self,
        image_shape: tuple[int, int],
        split_file: str,
        test_mode: bool,
        normalize: bool,
        augmentations_db: dict[str, float],
        resize_method: str,
        mini: float = 1.0,
        num_frames: int = 1,
        benchmark: bool = False,
        decode_fields: list[str] = ["image", "depth"],
        inplace_fields: list[str] = ["camera_params", "cam2w"],
        **kwargs,
    ) -> None:
        super().__init__(
            image_shape=image_shape,
            split_file=split_file,
            test_mode=test_mode,
            benchmark=benchmark,
            normalize=normalize,
            augmentations_db=augmentations_db,
            resize_method=resize_method,
            mini=mini,
            num_frames=num_frames,
            decode_fields=(
                decode_fields if not test_mode else [*decode_fields, "points"]
            ),
            inplace_fields=inplace_fields,
            **kwargs,
        )

    def pre_pipeline(self, results):
        results = super().pre_pipeline(results)
        results["dense"] = [True] * self.num_frames * self.num_copies
        results["quality"] = [1] * self.num_frames * self.num_copies
        return results


class ETH3DRMVD(SequenceDataset):
    min_depth = 0.01
    max_depth = 50.0
    depth_scale = 1000.0
    default_fps = 6
    test_split = "test.txt"
    train_split = "test.txt"
    sequences_file = "sequences.json"
    hdf5_paths = ["eth3d_rmvd.hdf5"]

    def __init__(
        self,
        image_shape,
        split_file,
        test_mode,
        crop=None,
        augmentations_db={},
        normalize=True,
        resize_method="hard",
        mini: float = 1.0,
        num_frames: int = 1,
        benchmark: bool = False,
        decode_fields: list[str] = ["image", "depth"],
        inplace_fields: list[str] = ["K", "cam2w"],
        **kwargs,
    ):
        super().__init__(
            image_shape=image_shape,
            split_file=split_file,
            test_mode=test_mode,
            benchmark=benchmark,
            normalize=normalize,
            augmentations_db=augmentations_db,
            resize_method=resize_method,
            mini=mini,
            num_frames=num_frames,
            decode_fields=decode_fields,
            inplace_fields=inplace_fields,
            **kwargs,
        )