Upload dropjects.py with huggingface_hub
Browse files- dropjects.py +110 -0
dropjects.py
ADDED
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import io
|
2 |
+
import itertools as it
|
3 |
+
import numpy as np
|
4 |
+
import datasets as d
|
5 |
+
|
6 |
+
|
7 |
+
_DESCRIPTION = """\
|
8 |
+
The Dropjects dataset was created at the Chair of Cyber-Physical Systems in Production \
|
9 |
+
Engineering at the Technical University of Munich.
|
10 |
+
"""
|
11 |
+
|
12 |
+
SUBSETS = [
|
13 |
+
"omni",
|
14 |
+
"cps",
|
15 |
+
"linemod",
|
16 |
+
"ycbv",
|
17 |
+
"homebreweddb",
|
18 |
+
"hope",
|
19 |
+
"tless",
|
20 |
+
]
|
21 |
+
|
22 |
+
NUM_SHARDS = {
|
23 |
+
"cps": 1000,
|
24 |
+
"ycbv": 1000,
|
25 |
+
"linemod": 1000,
|
26 |
+
"tless": 1000,
|
27 |
+
"omni": 10_000,
|
28 |
+
}
|
29 |
+
|
30 |
+
|
31 |
+
BASE_PATH = "https://huggingface.co/datasets/LukasDb/dropjects/resolve/main/data/train/{subset}/{shard}.tar"
|
32 |
+
|
33 |
+
|
34 |
+
h = 1440
|
35 |
+
w = 2560
|
36 |
+
|
37 |
+
|
38 |
+
class Dropjects(d.GeneratorBasedBuilder):
|
39 |
+
BUILDER_CONFIGS = list(d.BuilderConfig(name=x) for x in SUBSETS)
|
40 |
+
|
41 |
+
def _info(self):
|
42 |
+
|
43 |
+
features = d.Features(
|
44 |
+
{
|
45 |
+
# TODO at least the resolution is different
|
46 |
+
"rgb": d.Array3D((h, w, 3), dtype="uint8"),
|
47 |
+
"rgb_R": d.Array3D((h, w, 3), dtype="uint8"),
|
48 |
+
"depth": d.Array2D((h, w), dtype="float32"),
|
49 |
+
"depth_R": d.Array2D((h, w), dtype="float32"),
|
50 |
+
"mask": d.Array2D((h, w), dtype="int32"),
|
51 |
+
"obj_ids": d.Sequence(d.Value("int32")),
|
52 |
+
"obj_classes": d.Sequence(d.Value("string")),
|
53 |
+
"obj_pos": d.Sequence(d.Sequence(d.Value("float32"))),
|
54 |
+
"obj_rot": d.Sequence(d.Sequence(d.Value("float32"))),
|
55 |
+
"obj_bbox_obj": d.Sequence(d.Sequence(d.Value("int32"))),
|
56 |
+
"obj_bbox_visib": d.Sequence(d.Sequence(d.Value("int32"))),
|
57 |
+
"cam_location": d.Sequence(d.Value("float32")),
|
58 |
+
"cam_rotation": d.Sequence(d.Value("float32")),
|
59 |
+
"cam_matrix": d.Array2D((3, 3), dtype="float32"),
|
60 |
+
"obj_px_count_all": d.Sequence(d.Value("int32")),
|
61 |
+
"obj_px_count_valid": d.Sequence(d.Value("int32")),
|
62 |
+
"obj_px_count_visib": d.Sequence(d.Value("int32")),
|
63 |
+
"obj_visib_fract": d.Sequence(d.Value("float32")),
|
64 |
+
}
|
65 |
+
)
|
66 |
+
return d.DatasetInfo(
|
67 |
+
description=_DESCRIPTION,
|
68 |
+
citation="", # TODO
|
69 |
+
homepage="", # TODO
|
70 |
+
license="cc",
|
71 |
+
features=features,
|
72 |
+
)
|
73 |
+
|
74 |
+
def _split_generators(self, dl_manager):
|
75 |
+
subset = self.config.name
|
76 |
+
|
77 |
+
archive_paths = [
|
78 |
+
BASE_PATH.format(subset=subset, shard=i) for i in range(NUM_SHARDS[subset])
|
79 |
+
]
|
80 |
+
|
81 |
+
downloaded = dl_manager.download(archive_paths)
|
82 |
+
|
83 |
+
return [
|
84 |
+
d.SplitGenerator(
|
85 |
+
name=d.Split.TRAIN,
|
86 |
+
gen_kwargs={"tars": [dl_manager.iter_archive(d) for d in downloaded]},
|
87 |
+
),
|
88 |
+
]
|
89 |
+
|
90 |
+
def _generate_examples(self, tars):
|
91 |
+
sample = {}
|
92 |
+
id = None
|
93 |
+
|
94 |
+
for tar in tars:
|
95 |
+
for file_path, file_obj in tar:
|
96 |
+
new_id = file_path.split(".")[0]
|
97 |
+
if id is None:
|
98 |
+
id = new_id
|
99 |
+
else:
|
100 |
+
if id != new_id:
|
101 |
+
yield id, sample
|
102 |
+
sample = {}
|
103 |
+
id = new_id
|
104 |
+
|
105 |
+
key = file_path.split(".")[1]
|
106 |
+
|
107 |
+
bytes = io.BytesIO(file_obj.read())
|
108 |
+
value = np.load(bytes, allow_pickle=False)
|
109 |
+
|
110 |
+
sample[key] = value
|