File size: 8,146 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
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
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 DiodeIndoor(ImageDataset):
    CAM_INTRINSIC = {
        "ALL": torch.tensor([[886.81, 0, 512], [0, 927.06, 384], [0, 0, 1]])
    }
    min_depth = 0.01
    max_depth = 25.0
    depth_scale = 256.0
    test_split = "val.txt"
    train_split = "train.txt"
    hdf5_paths = ["DiodeIndoor.hdf5"]

    def __init__(
        self,
        image_shape,
        split_file,
        test_mode,
        crop=None,
        benchmark=False,
        augmentations_db={},
        normalize=True,
        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,
            mini=mini,
            **kwargs,
        )
        self.test_mode = test_mode

        # load annotations
        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
        h5file.close()
        dataset = []
        for line in txt_string.split("\n"):
            image_filename, depth_filename = line.strip().split(" ")
            sample = [
                image_filename,
                depth_filename,
            ]
            dataset.append(sample)

        if not self.test_mode:
            dataset = self.chunk(dataset, chunk_dim=1, pct=self.mini)

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

    def get_intrinsics(self, *args, **kwargs):
        return self.CAM_INTRINSIC["ALL"].clone()

    def get_mapper(self):
        return {
            "image_filename": 0,
            "depth_filename": 1,
        }

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


class DiodeIndoor_F(SequenceDataset):
    min_depth = 0.01
    max_depth = 25.0
    depth_scale = 1000.0
    test_split = "train.txt"
    train_split = "train.txt"
    sequences_file = "sequences.json"
    hdf5_paths = ["DiodeIndoor-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 DiodeOutdoor(ImageDataset):
    CAM_INTRINSIC = {
        "ALL": torch.tensor([[886.81, 0, 512], [0, 927.06, 384], [0, 0, 1]])
    }
    min_depth = 0.1
    max_depth = 80.0
    log_mean = 0
    log_std = 1
    test_split = "diode_outdoor_val.txt"
    train_split = "diode_outdoor_train.txt"
    hdf5_paths = ["diode.hdf5"]

    def __init__(
        self,
        image_shape,
        split_file,
        test_mode,
        depth_scale=256,
        crop=None,
        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.depth_scale = depth_scale

        self.masker = AnnotationMask(
            min_value=self.min_depth,
            max_value=self.max_depth if test_mode else None,
            custom_fn=self.eval_mask if test_mode else lambda x, *args, **kwargs: x,
        )
        # load annotations
        self.load_dataset()

    def load_dataset(self):
        self.h5file = h5py.File(
            os.path.join(self.data_root, self.hdf5_path),
            "r",
            libver="latest",
            swmr=True,
        )
        txt_file = np.array(self.h5file[self.split_file])
        txt_string = txt_file.tostring().decode("ascii")[:-1]
        dataset = {"depth_filename": [], "image_filename": []}
        for line in txt_string.split("\n"):
            depth_filename = line.strip().split(" ")[1]
            img_name = line.strip().split(" ")[0]
            image_filename = img_name
            dataset["depth_filename"].append(depth_filename)
            dataset["image_filename"].append(image_filename)

        self.dataset = pl.from_dict(dataset)

        if not self.test_mode and self.mini:
            self.dataset = self.dataset[::2]


class Diode(ImageDataset):
    CAM_INTRINSIC = {
        "ALL": torch.tensor([[886.81, 0, 512], [0, 927.06, 384], [0, 0, 1]])
    }
    log_mean = 0
    log_std = 1
    min_depth = 0.6
    max_depth = 80.0
    test_split = "diode_val.txt"
    train_split = "diode_train.txt"
    hdf5_paths = ["diode.hdf5"]

    def __init__(
        self,
        image_shape,
        split_file,
        test_mode,
        depth_scale=256,
        crop=None,
        benchmark=False,
        augmentations_db={},
        normalize=True,
        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,
            mini=mini,
            **kwargs,
        )
        self.test_mode = test_mode
        self.depth_scale = depth_scale

        self.masker = AnnotationMask(
            min_value=self.min_depth,
            max_value=self.max_depth if test_mode else None,
            custom_fn=self.eval_mask if test_mode else lambda x, *args, **kwargs: x,
        )
        # load annotations
        self.load_dataset()

    def load_dataset(self):
        self.h5file = h5py.File(
            os.path.join(self.data_root, self.hdf5_path),
            "r",
            libver="latest",
            swmr=True,
        )
        txt_file = np.array(self.h5file[self.split_file])
        txt_string = txt_file.tostring().decode("ascii")[:-1]
        dataset = {"depth_filename": [], "image_filename": []}
        for line in txt_string.split("\n"):
            depth_filename = line.strip().split(" ")[1]
            image_filename = line.strip().split(" ")[0]
            dataset["depth_filename"].append(depth_filename)
            dataset["image_filename"].append(image_filename)

        self.dataset = pl.from_dict(dataset)

        if not self.test_mode and self.mini:
            self.dataset = self.dataset[::2]

    def get_intrinsics(self, *args, **kwargs):
        return self.CAM_INTRINSIC["ALL"].clone()