parquet-converter commited on
Commit
047cd55
1 Parent(s): 525a71c

Update parquet files

Browse files
README.md DELETED
@@ -1,272 +0,0 @@
1
- ---
2
- annotations_creators:
3
- - machine-generated
4
- - expert-generated
5
- language_creators:
6
- - machine-generated
7
- language:
8
- - en
9
- license:
10
- - other
11
- multilinguality:
12
- - monolingual
13
- size_categories:
14
- - 100K<n<1M
15
- source_datasets:
16
- - original
17
- task_categories:
18
- - image-classification
19
- - object-detection
20
- task_ids: []
21
- paperswithcode_id: svhn
22
- pretty_name: Street View House Numbers
23
- dataset_info:
24
- - config_name: full_numbers
25
- features:
26
- - name: image
27
- dtype: image
28
- - name: digits
29
- sequence:
30
- - name: bbox
31
- sequence: int32
32
- length: 4
33
- - name: label
34
- dtype:
35
- class_label:
36
- names:
37
- 0: '0'
38
- 1: '1'
39
- 2: '2'
40
- 3: '3'
41
- 4: '4'
42
- 5: '5'
43
- 6: '6'
44
- 7: '7'
45
- 8: '8'
46
- 9: '9'
47
- splits:
48
- - name: train
49
- num_bytes: 390404309
50
- num_examples: 33402
51
- - name: test
52
- num_bytes: 271503052
53
- num_examples: 13068
54
- - name: extra
55
- num_bytes: 1868720340
56
- num_examples: 202353
57
- download_size: 2636187279
58
- dataset_size: 2530627701
59
- - config_name: cropped_digits
60
- features:
61
- - name: image
62
- dtype: image
63
- - name: label
64
- dtype:
65
- class_label:
66
- names:
67
- 0: '0'
68
- 1: '1'
69
- 2: '2'
70
- 3: '3'
71
- 4: '4'
72
- 5: '5'
73
- 6: '6'
74
- 7: '7'
75
- 8: '8'
76
- 9: '9'
77
- splits:
78
- - name: train
79
- num_bytes: 128364360
80
- num_examples: 73257
81
- - name: test
82
- num_bytes: 44464040
83
- num_examples: 26032
84
- - name: extra
85
- num_bytes: 967853504
86
- num_examples: 531131
87
- download_size: 1575594780
88
- dataset_size: 1140681904
89
- ---
90
-
91
- # Dataset Card for Street View House Numbers
92
-
93
- ## Table of Contents
94
- - [Table of Contents](#table-of-contents)
95
- - [Dataset Description](#dataset-description)
96
- - [Dataset Summary](#dataset-summary)
97
- - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
98
- - [Languages](#languages)
99
- - [Dataset Structure](#dataset-structure)
100
- - [Data Instances](#data-instances)
101
- - [Data Fields](#data-fields)
102
- - [Data Splits](#data-splits)
103
- - [Dataset Creation](#dataset-creation)
104
- - [Curation Rationale](#curation-rationale)
105
- - [Source Data](#source-data)
106
- - [Annotations](#annotations)
107
- - [Personal and Sensitive Information](#personal-and-sensitive-information)
108
- - [Considerations for Using the Data](#considerations-for-using-the-data)
109
- - [Social Impact of Dataset](#social-impact-of-dataset)
110
- - [Discussion of Biases](#discussion-of-biases)
111
- - [Other Known Limitations](#other-known-limitations)
112
- - [Additional Information](#additional-information)
113
- - [Dataset Curators](#dataset-curators)
114
- - [Licensing Information](#licensing-information)
115
- - [Citation Information](#citation-information)
116
- - [Contributions](#contributions)
117
-
118
- ## Dataset Description
119
-
120
- - **Homepage:** http://ufldl.stanford.edu/housenumbers
121
- - **Repository:**
122
- - **Paper:** [Reading Digits in Natural Images with Unsupervised Feature Learning](http://ufldl.stanford.edu/housenumbers/nips2011_housenumbers.pdf)
123
- - **Leaderboard:** https://paperswithcode.com/sota/image-classification-on-svhn
124
- - **Point of Contact:** streetviewhousenumbers@gmail.com
125
-
126
- ### Dataset Summary
127
-
128
- SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting. It can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images. The dataset comes in two formats:
129
- 1. Original images with character level bounding boxes.
130
- 2. MNIST-like 32-by-32 images centered around a single character (many of the images do contain some distractors at the sides).
131
-
132
- ### Supported Tasks and Leaderboards
133
-
134
- - `object-detection`: The dataset can be used to train a model for digit detection.
135
- - `image-classification`: The dataset can be used to train a model for Image Classification where the task is to predict a correct digit on the image. The leaderboard for this task is available at:
136
- https://paperswithcode.com/sota/image-classification-on-svhn
137
-
138
- ### Languages
139
-
140
- English
141
-
142
- ## Dataset Structure
143
-
144
- ### Data Instances
145
-
146
- #### full_numbers
147
-
148
- The original, variable-resolution, color house-number images with character level bounding boxes.
149
-
150
- ```
151
- {
152
- 'image': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=98x48 at 0x259E3F01780>,
153
- 'digits': {
154
- 'bbox': [
155
- [36, 7, 13, 32],
156
- [50, 7, 12, 32]
157
- ],
158
- 'label': [6, 9]
159
- }
160
- }
161
- ```
162
-
163
- #### cropped_digits
164
-
165
- Character level ground truth in an MNIST-like format. All digits have been resized to a fixed resolution of 32-by-32 pixels. The original character bounding boxes are extended in the appropriate dimension to become square windows, so that resizing them to 32-by-32 pixels does not introduce aspect ratio distortions. Nevertheless this preprocessing introduces some distracting digits to the sides of the digit of interest.
166
-
167
- ```
168
- {
169
- 'image': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=32x32 at 0x25A89494780>,
170
- 'label': 1
171
- }
172
- ```
173
-
174
- ### Data Fields
175
-
176
- #### full_numbers
177
-
178
- - `image`: A `PIL.Image.Image` object containing the image. Note that when accessing the image column: `dataset[0]["image"]` the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the `"image"` column, *i.e.* `dataset[0]["image"]` should **always** be preferred over `dataset["image"][0]`
179
- - `digits`: a dictionary containing digits' bounding boxes and labels
180
- - `bbox`: a list of bounding boxes (in the [coco](https://albumentations.ai/docs/getting_started/bounding_boxes_augmentation/#coco) format) corresponding to the digits present on the image
181
- - `label`: a list of integers between 0 and 9 representing the digit.
182
-
183
- #### cropped_digits
184
-
185
- - `image`: A `PIL.Image.Image` object containing the image. Note that when accessing the image column: `dataset[0]["image"]` the image file is automatically decoded. Decoding of a large number of image files might take a significant amount of time. Thus it is important to first query the sample index before the `"image"` column, *i.e.* `dataset[0]["image"]` should **always** be preferred over `dataset["image"][0]`
186
- - `digit`: an integer between 0 and 9 representing the digit.
187
-
188
- ### Data Splits
189
-
190
- #### full_numbers
191
-
192
- The data is split into training, test and extra set. The training set contains 33402 images, test set 13068 and the extra set 202353 images.
193
-
194
- #### cropped_digits
195
-
196
- The data is split into training, test and extra set. The training set contains 73257 images, test set 26032 and the extra set 531131 images.
197
-
198
- The extra set can be used as extra training data. The extra set was obtained in a similar manner to the training and test set, but with the increased detection threshold in order to generate this large amount of labeled data. The SVHN extra subset is thus somewhat biased toward less difficult detections, and is thus easier than SVHN train/SVHN test.
199
-
200
- ## Dataset Creation
201
-
202
- ### Curation Rationale
203
-
204
- From the paper:
205
- > As mentioned above, the venerable MNIST dataset has been a valuable goal post for researchers seeking to build better learning systems whose benchmark performance could be expected to translate into improved performance on realistic applications. However, computers have now reached essentially human levels of performance on this problem—a testament to progress in machine learning and computer vision. The Street View House Numbers (SVHN) digit database that we provide can be seen as similar in flavor to MNIST (e.g., the images are of small cropped characters), but the SVHN dataset incorporates an order of magnitude more labeled data and comes from a significantly harder, unsolved, real world problem. Here the gap between human performance and state of the art feature representations is significant. Going forward, we expect that this dataset may fulfill a similar role for modern feature learning algorithms: it provides a new and difficult benchmark where increased performance can be expected to translate into tangible gains on a realistic application.
206
-
207
- ### Source Data
208
-
209
- #### Initial Data Collection and Normalization
210
-
211
- From the paper:
212
- > The SVHN dataset was obtained from a large number of Street View images using a combination
213
- of automated algorithms and the Amazon Mechanical Turk (AMT) framework, which was
214
- used to localize and transcribe the single digits. We downloaded a very large set of images from
215
- urban areas in various countries.
216
-
217
- #### Who are the source language producers?
218
-
219
- [More Information Needed]
220
-
221
- ### Annotations
222
-
223
- #### Annotation process
224
-
225
- From the paper:
226
- > From these randomly selected images, the house-number patches were extracted using a dedicated sliding window house-numbers detector using a low threshold on the detector’s confidence score in order to get a varied, unbiased dataset of house-number signs. These low precision detections were screened and transcribed by AMT workers.
227
-
228
- #### Who are the annotators?
229
-
230
- The AMT workers.
231
-
232
- ### Personal and Sensitive Information
233
-
234
- [More Information Needed]
235
-
236
- ## Considerations for Using the Data
237
-
238
- ### Social Impact of Dataset
239
-
240
- [More Information Needed]
241
-
242
- ### Discussion of Biases
243
-
244
- [More Information Needed]
245
-
246
- ### Other Known Limitations
247
-
248
- [More Information Needed]
249
-
250
- ## Additional Information
251
-
252
- ### Dataset Curators
253
-
254
- Yuval Netzer, Tao Wang, Adam Coates, Alessandro Bissacco, Bo Wu and Andrew Y. Ng
255
-
256
- ### Licensing Information
257
-
258
- Non-commerical use only.
259
-
260
- ### Citation Information
261
-
262
- ```
263
- @article{netzer2011reading,
264
- title={Reading digits in natural images with unsupervised feature learning},
265
- author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Bo and Ng, Andrew Y},
266
- year={2011}
267
- }
268
- ```
269
-
270
- ### Contributions
271
-
272
- Thanks to [@mariosasko](https://github.com/mariosasko) for adding this dataset.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cropped_digits/svhn-extra-00000-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39a962c367f4f0c8c7dd604e09a0853d984741b7a01a7f15a95ee48d613ce48a
3
+ size 501213476
cropped_digits/svhn-extra-00001-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:31b18a34da0c622bb9ff21509c874638d71fedcf2d1379d3da1de023aa13463a
3
+ size 467018470
cropped_digits/svhn-test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e93c2f59077400ddde8c793f28612a540aa2f622f93cccf75b5fe5c5be6a0f2e
3
+ size 44477871
cropped_digits/svhn-train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c248ddffda6903a5bb590ab3f8d40748a358c832354a13659f4e83885278c86
3
+ size 128387775
dataset_infos.json DELETED
@@ -1 +0,0 @@
1
- {"full_numbers": {"description": "SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting.\nIt can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images)\nand comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.\n", "citation": "@article{netzer2011reading,\n title={Reading digits in natural images with unsupervised feature learning},\n author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Bo and Ng, Andrew Y},\n year={2011}\n}\n", "homepage": "http://ufldl.stanford.edu/housenumbers/", "license": "Custom (non-commercial)", "features": {"image": {"id": null, "_type": "Image"}, "digits": {"feature": {"bbox": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": 4, "id": null, "_type": "Sequence"}, "label": {"num_classes": 10, "names": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "svhn", "config_name": "full_numbers", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 390404309, "num_examples": 33402, "dataset_name": "svhn"}, "test": {"name": "test", "num_bytes": 271503052, "num_examples": 13068, "dataset_name": "svhn"}, "extra": {"name": "extra", "num_bytes": 1868720340, "num_examples": 202353, "dataset_name": "svhn"}}, "download_checksums": {"http://ufldl.stanford.edu/housenumbers/train.tar.gz": {"num_bytes": 404141560, "checksum": "4b17bb33b6cd8f963493168f80143da956f28ec406cc12f8e5745a9f91a51898"}, "http://ufldl.stanford.edu/housenumbers/test.tar.gz": {"num_bytes": 276555967, "checksum": "57ac9ceb530e4aa85b55d991be8fc49c695b3d71c6f6a88afea86549efde7fb5"}, "http://ufldl.stanford.edu/housenumbers/extra.tar.gz": {"num_bytes": 1955489752, "checksum": "e857e27d1e65bd1e7d3959b094061777f6506bbc39889a0df3bba6a729d60f9c"}}, "download_size": 2636187279, "post_processing_size": null, "dataset_size": 2530627701, "size_in_bytes": 5166814980}, "cropped_digits": {"description": "SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting.\nIt can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images)\nand comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.\n", "citation": "@article{netzer2011reading,\n title={Reading digits in natural images with unsupervised feature learning},\n author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Bo and Ng, Andrew Y},\n year={2011}\n}\n", "homepage": "http://ufldl.stanford.edu/housenumbers/", "license": "Custom (non-commercial)", "features": {"image": {"id": null, "_type": "Image"}, "label": {"num_classes": 10, "names": ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": [{"task": "image-classification", "image_column": "image", "label_column": "label", "labels": null}], "builder_name": "svhn", "config_name": "cropped_digits", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 128364360, "num_examples": 73257, "dataset_name": "svhn"}, "test": {"name": "test", "num_bytes": 44464040, "num_examples": 26032, "dataset_name": "svhn"}, "extra": {"name": "extra", "num_bytes": 967853504, "num_examples": 531131, "dataset_name": "svhn"}}, "download_checksums": {"http://ufldl.stanford.edu/housenumbers/train_32x32.mat": {"num_bytes": 182040794, "checksum": "435e94d69a87fde4fd4d7f3dd208dfc32cb6ae8af2240d066de1df7508d083b8"}, "http://ufldl.stanford.edu/housenumbers/test_32x32.mat": {"num_bytes": 64275384, "checksum": "cdce80dfb2a2c4c6160906d0bd7c68ec5a99d7ca4831afa54f09182025b6a75b"}, "http://ufldl.stanford.edu/housenumbers/extra_32x32.mat": {"num_bytes": 1329278602, "checksum": "a133a4beb38a00fcdda90c9489e0c04f900b660ce8a316a5e854838379a71eb3"}}, "download_size": 1575594780, "post_processing_size": null, "dataset_size": 1140681904, "size_in_bytes": 2716276684}}
 
 
full_numbers/svhn-extra-00000-of-00004.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0aacd7b3688603a8a4e39e1fd409a68a995ab3b6f94f04b66e4eabffe1c02b5e
3
+ size 497604227
full_numbers/svhn-extra-00001-of-00004.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:277ecc4fd28911bc5319c9634c132a64e359bce22423a70f63e1980e033ad5f8
3
+ size 503362262
full_numbers/svhn-extra-00002-of-00004.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3de636c09faaf4648af4d6197da0fed4b753a630386e3aa0a5a07c41235cc078
3
+ size 505390918
full_numbers/svhn-extra-00003-of-00004.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:56dac711ce2360c9eed6961758d79a38a24dac6baf695443b14a15ac32a6299d
3
+ size 349660674
full_numbers/svhn-test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c27f648f2fadb638d0914e6de0fab9cc8c3568f277afe58aa69589563e87cdcd
3
+ size 270881696
full_numbers/svhn-train.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3de4f181147d45dff9a13e780647c434288571b423fa0fa5d275458c668a2570
3
+ size 388561950
svhn.py DELETED
@@ -1,199 +0,0 @@
1
- # coding=utf-8
2
- # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
3
- #
4
- # Licensed under the Apache License, Version 2.0 (the "License");
5
- # you may not use this file except in compliance with the License.
6
- # You may obtain a copy of the License at
7
- #
8
- # http://www.apache.org/licenses/LICENSE-2.0
9
- #
10
- # Unless required by applicable law or agreed to in writing, software
11
- # distributed under the License is distributed on an "AS IS" BASIS,
12
- # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
- # See the License for the specific language governing permissions and
14
- # limitations under the License.
15
- """Street View House Numbers (SVHN) dataset."""
16
-
17
- import io
18
- import os
19
-
20
- import h5py
21
- import numpy as np
22
- import scipy.io as sio
23
-
24
- import datasets
25
- from datasets.tasks import ImageClassification
26
-
27
-
28
- logger = datasets.logging.get_logger(__name__)
29
-
30
-
31
- _CITATION = """\
32
- @article{netzer2011reading,
33
- title={Reading digits in natural images with unsupervised feature learning},
34
- author={Netzer, Yuval and Wang, Tao and Coates, Adam and Bissacco, Alessandro and Wu, Bo and Ng, Andrew Y},
35
- year={2011}
36
- }
37
- """
38
-
39
- _DESCRIPTION = """\
40
- SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting.
41
- It can be seen as similar in flavor to MNIST (e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images)
42
- and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images.
43
- """
44
-
45
- _HOMEPAGE = "http://ufldl.stanford.edu/housenumbers/"
46
-
47
- _LICENSE = "Custom (non-commercial)"
48
-
49
- _URLs = {
50
- "full_numbers": [
51
- "http://ufldl.stanford.edu/housenumbers/train.tar.gz",
52
- "http://ufldl.stanford.edu/housenumbers/test.tar.gz",
53
- "http://ufldl.stanford.edu/housenumbers/extra.tar.gz",
54
- ],
55
- "cropped_digits": [
56
- "http://ufldl.stanford.edu/housenumbers/train_32x32.mat",
57
- "http://ufldl.stanford.edu/housenumbers/test_32x32.mat",
58
- "http://ufldl.stanford.edu/housenumbers/extra_32x32.mat",
59
- ],
60
- }
61
-
62
- _DIGIT_LABELS = [str(num) for num in range(10)]
63
-
64
-
65
- class SVHN(datasets.GeneratorBasedBuilder):
66
- """Street View House Numbers (SVHN) dataset."""
67
-
68
- VERSION = datasets.Version("1.0.0")
69
-
70
- BUILDER_CONFIGS = [
71
- datasets.BuilderConfig(
72
- name="full_numbers",
73
- version=VERSION,
74
- description="Contains the original, variable-resolution, color house-number images with character level bounding boxes.",
75
- ),
76
- datasets.BuilderConfig(
77
- name="cropped_digits",
78
- version=VERSION,
79
- description="Character level ground truth in an MNIST-like format. All digits have been resized to a fixed resolution of 32-by-32 pixels. The original character bounding boxes are extended in the appropriate dimension to become square windows, so that resizing them to 32-by-32 pixels does not introduce aspect ratio distortions. Nevertheless this preprocessing introduces some distracting digits to the sides of the digit of interest.",
80
- ),
81
- ]
82
-
83
- def _info(self):
84
- if self.config.name == "full_numbers":
85
- features = datasets.Features(
86
- {
87
- "image": datasets.Image(),
88
- "digits": datasets.Sequence(
89
- {
90
- "bbox": datasets.Sequence(datasets.Value("int32"), length=4),
91
- "label": datasets.ClassLabel(num_classes=10),
92
- }
93
- ),
94
- }
95
- )
96
- else:
97
- features = datasets.Features(
98
- {
99
- "image": datasets.Image(),
100
- "label": datasets.ClassLabel(num_classes=10),
101
- }
102
- )
103
- return datasets.DatasetInfo(
104
- description=_DESCRIPTION,
105
- features=features,
106
- supervised_keys=None,
107
- homepage=_HOMEPAGE,
108
- license=_LICENSE,
109
- citation=_CITATION,
110
- task_templates=[ImageClassification(image_column="image", label_column="label")]
111
- if self.config.name == "cropped_digits"
112
- else None,
113
- )
114
-
115
- def _split_generators(self, dl_manager):
116
- if self.config.name == "full_numbers":
117
- train_archive, test_archive, extra_archive = dl_manager.download(_URLs[self.config.name])
118
- for path, f in dl_manager.iter_archive(train_archive):
119
- if path.endswith("digitStruct.mat"):
120
- train_annot_data = f.read()
121
- break
122
- for path, f in dl_manager.iter_archive(test_archive):
123
- if path.endswith("digitStruct.mat"):
124
- test_annot_data = f.read()
125
- break
126
- for path, f in dl_manager.iter_archive(extra_archive):
127
- if path.endswith("digitStruct.mat"):
128
- extra_annot_data = f.read()
129
- break
130
- train_archive = dl_manager.iter_archive(train_archive)
131
- test_archive = dl_manager.iter_archive(test_archive)
132
- extra_archive = dl_manager.iter_archive(extra_archive)
133
- train_filepath, test_filepath, extra_filepath = None, None, None
134
- else:
135
- train_annot_data, test_annot_data, extra_annot_data = None, None, None
136
- train_archive, test_archive, extra_archive = None, None, None
137
- train_filepath, test_filepath, extra_filepath = dl_manager.download(_URLs[self.config.name])
138
- return [
139
- datasets.SplitGenerator(
140
- name=datasets.Split.TRAIN,
141
- gen_kwargs={
142
- "annot_data": train_annot_data,
143
- "files": train_archive,
144
- "filepath": train_filepath,
145
- },
146
- ),
147
- datasets.SplitGenerator(
148
- name=datasets.Split.TEST,
149
- gen_kwargs={
150
- "annot_data": test_annot_data,
151
- "files": test_archive,
152
- "filepath": test_filepath,
153
- },
154
- ),
155
- datasets.SplitGenerator(
156
- name="extra",
157
- gen_kwargs={
158
- "annot_data": extra_annot_data,
159
- "files": extra_archive,
160
- "filepath": extra_filepath,
161
- },
162
- ),
163
- ]
164
-
165
- def _generate_examples(self, annot_data, files, filepath):
166
- if self.config.name == "full_numbers":
167
-
168
- def _get_digits(bboxes, h5_file):
169
- def key_to_values(key, bbox):
170
- if bbox[key].shape[0] == 1:
171
- return [int(bbox[key][0][0])]
172
- else:
173
- return [int(h5_file[bbox[key][i][0]][()].item()) for i in range(bbox[key].shape[0])]
174
-
175
- bbox = h5_file[bboxes[0]]
176
- assert bbox.keys() == {"height", "left", "top", "width", "label"}
177
- bbox_columns = [key_to_values(key, bbox) for key in ["left", "top", "width", "height", "label"]]
178
- return [
179
- {"bbox": [left, top, width, height], "label": label % 10}
180
- for left, top, width, height, label in zip(*bbox_columns)
181
- ]
182
-
183
- with h5py.File(io.BytesIO(annot_data), "r") as h5_file:
184
- for path, f in files:
185
- root, ext = os.path.splitext(path)
186
- if ext != ".png":
187
- continue
188
- img_idx = int(os.path.basename(root)) - 1
189
- yield img_idx, {
190
- "image": {"path": path, "bytes": f.read()},
191
- "digits": _get_digits(h5_file["digitStruct/bbox"][img_idx], h5_file),
192
- }
193
- else:
194
- data = sio.loadmat(filepath)
195
- for i, (image_array, label) in enumerate(zip(np.rollaxis(data["X"], -1), data["y"])):
196
- yield i, {
197
- "image": image_array,
198
- "label": label.item() % 10,
199
- }