-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
194 lines (144 loc) · 8.09 KB
/
dataset.py
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
from base_dataset import AnnDataset, AnnDatasetSelfTrain
class Sift1M(AnnDatasetSelfTrain):
# http://corpus-texmex.irisa.fr/
# {'nb': 1000000, 'dim': 128, 'train_queries': (100000, 128), 'train_gts': (100000, 100), 'test_queries': (10000, 128), 'test_gts': (10000, 100), 'self_train_gts': (1000000, 100)}
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../sift1M") -> None:
super().__init__("sift-1M", "l2", path, train_set_len, self_train_set_len)
self.base_fn = 'sift_base.fvecs'
self.train_query_fn = 'sift_learn.fvecs'
self.train_gt_fn = 'sift-1M.learn_l2.100K.ivecs'
self.test_query_fn = 'sift_query.fvecs'
self.test_gt_fn = 'sift_groundtruth.ivecs'
self.self_train_gt_fn = 'sift-1M.self_learn_l2.1M.ivecs'
class Text2image10M(AnnDatasetSelfTrain):
# image unnormed
# https://research.yandex.com/blog/benchmarks-for-billion-scale-similarity-search
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../text2image") -> None:
super().__init__("Text2Image-10M", "ip", path, train_set_len, self_train_set_len)
if train_set_len < 0:
self.train_set_len = 10000000
self.base_fn = 'base.1B.fbin.crop_nb_10000000.fbin'
self.train_query_fn = 'query.learn.50M.fbin'
self.train_gt_fn = 'text2image-10M.learn_ip.10M.ivecs'
self.test_query_fn = 'query.public.10K.fbin'
self.test_gt_fn = 'text2image-10M.10K.ibin'
self.self_test_query_fn = 'self_query.public.10K.fbin'
self.self_test_gt_fn = 'Text2Image-10M.self_test_ip.ivecs'
self.self_train_gt_fn = 'text2image-10M.self_ip.ivecs'
class Text2image100M(AnnDatasetSelfTrain):
# image unnormed
# https://research.yandex.com/blog/benchmarks-for-billion-scale-similarity-search
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../Text2image100M") -> None:
super().__init__("Text2Image-100M", "ip", path, train_set_len, self_train_set_len)
# if train_set_len < 0:
# self.train_set_len = 10000000
self.base_fn = 'base.100M.fbin'
# self.train_query_fn = 'query.learn.50M.fbin'
# self.train_gt_fn = 'text2image-10M.learn_ip.10M.ivecs'
self.test_query_fn = 'query.public.100K.fbin'
self.test_gt_fn = 'text2image-10M.ibin'
# self.self_train_gt_fn = 'text2image-10M.self_ip.ivecs'
class Deep100M(AnnDataset):
# https://research.yandex.com/blog/benchmarks-for-billion-scale-similarity-search
# info = {'nb': 100000000, 'dim': 96, 'train_queries': (10000000, 96), 'train_gts': (10000000, 100), 'test_queries': (10000, 96), 'test_gts': (10000, 100), 'self_train_gts': (10000000, 100)}
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../deep") -> None:
super().__init__("deep-100M","l2", path, train_set_len)
self.base_fn = 'base.100M.fbin'
self.train_query_fn = 'learn.10M.fbin'
self.train_gt_fn = 'deep100M_gt.learn.10M.ivecs'
self.test_query_fn = 'query.public.10K.fbin'
self.test_gt_fn = 'deep100M_groundtruth.ivecs'
self.self_train_gt_fn = 'deep100M_gt.self_learn.10M.ivecs'
class Webvid(AnnDatasetSelfTrain):
# https://zenodo.org/records/11090378
# {'nb': 2505000, 'dim': 512, 'train_queries': (2500000, 512), 'train_gts': (2500000, 100), 'test_queries': (10000, 512), 'test_gts': (10000, 100), 'self_train_gts': (2505000, 100)}
# OOD
# l2-normalized
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../webvid_split") -> None:
super().__init__("webvid-2.5M", "ip", path, train_set_len, self_train_set_len)
self.base_fn = 'clip.webvid.base.2.5M.fbin' # Video
self.train_query_fn = 'webvid.query.train.2.5M.fbin' # Text
self.train_gt_fn = 'webvid-2.5M.learn_ip.2.5M.ibin'
self.test_query_fn = 'webvid.query.10k.fbin' # Text
self.test_gt_fn = 'webvid-2.5M.10k.ibin'
self.self_test_query_fn = 'self_webvid.query.10k.fbin'
self.self_test_gt_fn = 'self_webvid-2.5M.self_learn_ip.2.5M.ibin'
self.self_train_gt_fn = 'webvid-2.5M.self_learn_ip.2.5M.ibin'
class Laion(AnnDatasetSelfTrain):
# https://zenodo.org/records/11090378
# OOD
# l2-normalized
# {'nb': 10004480, 'dim': 512, 'train_queries': (1000000, 512), 'train_gts': (1000000, 100), 'test_queries': (10000, 512), 'test_gts': (10000, 100), 'self_train_gts': (10004480, 100)}
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../laion-10M_split") -> None:
super().__init__("laion-10M", "ip", path, train_set_len, self_train_set_len)
self.base_fn = 'base.10M.fbin' # Img
self.train_query_fn = 'query.train.10M.fbin' # Text
self.train_gt_fn = 'laion-10M.learn_ip.10M.ibin'
self.test_query_fn = 'laion.query.10k.fbin' # Text
# self.test_gt_fn = 'laion.gt.10k.ibin'
self.test_gt_fn = 'laion.gt_ip.10k.ibin'
self.self_test_query_fn = 'self_laion.query.10k.fbin'
self.self_test_gt_fn = 'self_laion-10M.self_learn_ip.10M.ibin'
self.self_train_gt_fn = 'laion-10M.self_learn_ip.10M.ibin'
class Websearch(AnnDataset):
# https://github.com/microsoft/MS-MARCO-Web-Search
def __init__(self, train_set_len = -1, self_train_set_len = -1, path = "../msmacro") -> None:
super().__init__("Websearch-100M", "ip", path, train_set_len)
self.base_fn = 'vectors.bin' # Doc Text
# self.train_query_fn = 'query.train.10M.fbin' # Text
# self.train_gt_fn = 'laion-10M.learn_ip.10M.ivecs'
self.test_query_fn = 'vectors_query.bin' # Query Text
self.test_gt_fn = 'msmacro100m.gt.ibin'
dataset_dict = { 'Deep100M':Deep100M,'Laion':Laion,'Sift1M':Sift1M,'Text2image10M':Text2image10M,'Websearch':Websearch,'Webvid':Webvid,'Text2image100M':Text2image100M }
def dataset_factory(dataset_name, train_set_len = -1, self_train_set_len = -1, read_mode = "only_test"):
if dataset_name in dataset_dict:
ds_class = dataset_dict[dataset_name]
if issubclass(ds_class, AnnDatasetSelfTrain):
ds:AnnDatasetSelfTrain = ds_class(train_set_len=train_set_len, self_train_set_len = self_train_set_len)
else:
ds:AnnDataset = ds_class(train_set_len=train_set_len)
else:
raise NotImplementedError(dataset_name)
if read_mode == "no_read":
return ds
if read_mode == "only_vecs":
ds.read_vecs()
elif read_mode == "only_test":
if issubclass(ds_class, AnnDatasetSelfTrain):
ds.read(load_train = False, load_self_train = False)
else:
ds.read(False)
elif read_mode == "all":
ds.read()
elif read_mode == "query":
if issubclass(ds_class, AnnDatasetSelfTrain):
ds.read(load_train = True, load_self_train = False)
else:
ds.read()
elif read_mode == "self":
assert issubclass(ds_class, AnnDatasetSelfTrain)
ds.read(load_train = False, load_self_train = True)
else:
raise NotImplementedError
return ds
if __name__=="__main__":
import inspect
import argparse
parser = argparse.ArgumentParser()
parser.add_argument("--getdict", "-g", action="store_true")
parser.add_argument("-d","--dataset", type=str, default="")
parser.add_argument("-i","--info", action="store_true")
args = parser.parse_args()
if args.getdict:
classes = []
for name, member in inspect.getmembers(__import__(__name__)):
if inspect.isclass(member):
classes.append(name)
dataset_dict[name] = member
print('{', ",".join([f"'{x}':{x}" for x in classes]) ,"}")
if args.dataset != "":
ds = dataset_dict[args.dataset](1000000)
print(" ".join(ds.files()))
if args.info:
ds.read()
print(ds.info())