Hugging Face's logo

Datasets:
WNJXYK
/
TTA-ImageNet-C

Tasks:
Image Classification
Modalities:
Image
Formats:
parquet
Size:
1M - 10M
Tags:
imagenet
corruption
robustness
tta
test-time-adaptation
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
WNJXYK commited on
Commit
bb0fa9d
·
verified ·
1 Parent(s): f78a27b

Update dataset card

Browse files
Files changed (1) hide show
  1. README.md +232 -0
README.md ADDED
@@ -0,0 +1,232 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ task_categories: [image-classification]
4
+ tags: [imagenet, corruption, robustness, tta, test-time-adaptation]
5
+ pretty_name: "TTA ImageNet-C"
6
+ size_categories: [1M<n<10M]
7
+ configs:
8
+ - config_name: gaussian_noise
9
+ data_files:
10
+ - split: severity_1
11
+ path: data/gaussian_noise/severity_1/*.parquet
12
+ - split: severity_2
13
+ path: data/gaussian_noise/severity_2/*.parquet
14
+ - split: severity_3
15
+ path: data/gaussian_noise/severity_3/*.parquet
16
+ - split: severity_4
17
+ path: data/gaussian_noise/severity_4/*.parquet
18
+ - split: severity_5
19
+ path: data/gaussian_noise/severity_5/*.parquet
20
+ - config_name: shot_noise
21
+ data_files:
22
+ - split: severity_1
23
+ path: data/shot_noise/severity_1/*.parquet
24
+ - split: severity_2
25
+ path: data/shot_noise/severity_2/*.parquet
26
+ - split: severity_3
27
+ path: data/shot_noise/severity_3/*.parquet
28
+ - split: severity_4
29
+ path: data/shot_noise/severity_4/*.parquet
30
+ - split: severity_5
31
+ path: data/shot_noise/severity_5/*.parquet
32
+ - config_name: impulse_noise
33
+ data_files:
34
+ - split: severity_1
35
+ path: data/impulse_noise/severity_1/*.parquet
36
+ - split: severity_2
37
+ path: data/impulse_noise/severity_2/*.parquet
38
+ - split: severity_3
39
+ path: data/impulse_noise/severity_3/*.parquet
40
+ - split: severity_4
41
+ path: data/impulse_noise/severity_4/*.parquet
42
+ - split: severity_5
43
+ path: data/impulse_noise/severity_5/*.parquet
44
+ - config_name: defocus_blur
45
+ data_files:
46
+ - split: severity_1
47
+ path: data/defocus_blur/severity_1/*.parquet
48
+ - split: severity_2
49
+ path: data/defocus_blur/severity_2/*.parquet
50
+ - split: severity_3
51
+ path: data/defocus_blur/severity_3/*.parquet
52
+ - split: severity_4
53
+ path: data/defocus_blur/severity_4/*.parquet
54
+ - split: severity_5
55
+ path: data/defocus_blur/severity_5/*.parquet
56
+ - config_name: glass_blur
57
+ data_files:
58
+ - split: severity_1
59
+ path: data/glass_blur/severity_1/*.parquet
60
+ - split: severity_2
61
+ path: data/glass_blur/severity_2/*.parquet
62
+ - split: severity_3
63
+ path: data/glass_blur/severity_3/*.parquet
64
+ - split: severity_4
65
+ path: data/glass_blur/severity_4/*.parquet
66
+ - split: severity_5
67
+ path: data/glass_blur/severity_5/*.parquet
68
+ - config_name: motion_blur
69
+ data_files:
70
+ - split: severity_1
71
+ path: data/motion_blur/severity_1/*.parquet
72
+ - split: severity_2
73
+ path: data/motion_blur/severity_2/*.parquet
74
+ - split: severity_3
75
+ path: data/motion_blur/severity_3/*.parquet
76
+ - split: severity_4
77
+ path: data/motion_blur/severity_4/*.parquet
78
+ - split: severity_5
79
+ path: data/motion_blur/severity_5/*.parquet
80
+ - config_name: zoom_blur
81
+ data_files:
82
+ - split: severity_1
83
+ path: data/zoom_blur/severity_1/*.parquet
84
+ - split: severity_2
85
+ path: data/zoom_blur/severity_2/*.parquet
86
+ - split: severity_3
87
+ path: data/zoom_blur/severity_3/*.parquet
88
+ - split: severity_4
89
+ path: data/zoom_blur/severity_4/*.parquet
90
+ - split: severity_5
91
+ path: data/zoom_blur/severity_5/*.parquet
92
+ - config_name: snow
93
+ data_files:
94
+ - split: severity_1
95
+ path: data/snow/severity_1/*.parquet
96
+ - split: severity_2
97
+ path: data/snow/severity_2/*.parquet
98
+ - split: severity_3
99
+ path: data/snow/severity_3/*.parquet
100
+ - split: severity_4
101
+ path: data/snow/severity_4/*.parquet
102
+ - split: severity_5
103
+ path: data/snow/severity_5/*.parquet
104
+ - config_name: frost
105
+ data_files:
106
+ - split: severity_1
107
+ path: data/frost/severity_1/*.parquet
108
+ - split: severity_2
109
+ path: data/frost/severity_2/*.parquet
110
+ - split: severity_3
111
+ path: data/frost/severity_3/*.parquet
112
+ - split: severity_4
113
+ path: data/frost/severity_4/*.parquet
114
+ - split: severity_5
115
+ path: data/frost/severity_5/*.parquet
116
+ - config_name: fog
117
+ data_files:
118
+ - split: severity_1
119
+ path: data/fog/severity_1/*.parquet
120
+ - split: severity_2
121
+ path: data/fog/severity_2/*.parquet
122
+ - split: severity_3
123
+ path: data/fog/severity_3/*.parquet
124
+ - split: severity_4
125
+ path: data/fog/severity_4/*.parquet
126
+ - split: severity_5
127
+ path: data/fog/severity_5/*.parquet
128
+ - config_name: brightness
129
+ data_files:
130
+ - split: severity_1
131
+ path: data/brightness/severity_1/*.parquet
132
+ - split: severity_2
133
+ path: data/brightness/severity_2/*.parquet
134
+ - split: severity_3
135
+ path: data/brightness/severity_3/*.parquet
136
+ - split: severity_4
137
+ path: data/brightness/severity_4/*.parquet
138
+ - split: severity_5
139
+ path: data/brightness/severity_5/*.parquet
140
+ - config_name: contrast
141
+ data_files:
142
+ - split: severity_1
143
+ path: data/contrast/severity_1/*.parquet
144
+ - split: severity_2
145
+ path: data/contrast/severity_2/*.parquet
146
+ - split: severity_3
147
+ path: data/contrast/severity_3/*.parquet
148
+ - split: severity_4
149
+ path: data/contrast/severity_4/*.parquet
150
+ - split: severity_5
151
+ path: data/contrast/severity_5/*.parquet
152
+ - config_name: elastic_transform
153
+ data_files:
154
+ - split: severity_1
155
+ path: data/elastic_transform/severity_1/*.parquet
156
+ - split: severity_2
157
+ path: data/elastic_transform/severity_2/*.parquet
158
+ - split: severity_3
159
+ path: data/elastic_transform/severity_3/*.parquet
160
+ - split: severity_4
161
+ path: data/elastic_transform/severity_4/*.parquet
162
+ - split: severity_5
163
+ path: data/elastic_transform/severity_5/*.parquet
164
+ - config_name: pixelate
165
+ data_files:
166
+ - split: severity_1
167
+ path: data/pixelate/severity_1/*.parquet
168
+ - split: severity_2
169
+ path: data/pixelate/severity_2/*.parquet
170
+ - split: severity_3
171
+ path: data/pixelate/severity_3/*.parquet
172
+ - split: severity_4
173
+ path: data/pixelate/severity_4/*.parquet
174
+ - split: severity_5
175
+ path: data/pixelate/severity_5/*.parquet
176
+ - config_name: jpeg_compression
177
+ data_files:
178
+ - split: severity_1
179
+ path: data/jpeg_compression/severity_1/*.parquet
180
+ - split: severity_2
181
+ path: data/jpeg_compression/severity_2/*.parquet
182
+ - split: severity_3
183
+ path: data/jpeg_compression/severity_3/*.parquet
184
+ - split: severity_4
185
+ path: data/jpeg_compression/severity_4/*.parquet
186
+ - split: severity_5
187
+ path: data/jpeg_compression/severity_5/*.parquet
188
+ ---
189
+
190
+ # TTA-ImageNet-C
191
+
192
+ Mirror of **ImageNet-C** (Hendrycks & Dietterich, ICLR 2019) with a revision
193
+ pin for reproducible test-time adaptation evaluation.
194
+
195
+ - **Upstream**: [Zenodo record 2235448](https://zenodo.org/records/2235448)
196
+ - **License**: CC BY 4.0 (matches upstream)
197
+ - **Maintained as part of**: [TTA-Evaluation-Harness](https://github.com/WNJXYK/TTA-Evaluation-Harness)
198
+
199
+ ## Citation
200
+
201
+ ```bibtex
202
+ @inproceedings{hendrycks2019benchmarking,
203
+ title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},
204
+ author={Hendrycks, Dan and Dietterich, Thomas},
205
+ booktitle={ICLR},
206
+ year={2019}
207
+ }
208
+ ```
209
+
210
+ ## Structure
211
+
212
+ - **15 configs**: one per corruption type (`gaussian_noise`, `shot_noise`, ..., `jpeg_compression`).
213
+ - **5 splits per config**: `severity_1` through `severity_5`, 50 000 images each (1000 classes x 50).
214
+ - Labels are `ClassLabel` with 1000 WordNet-ID names in torchvision order
215
+ (lexicographic on wnid; `n01440764` = idx 0 = tench).
216
+
217
+ ## Usage
218
+
219
+ ```python
220
+ from datasets import load_dataset
221
+
222
+ ds = load_dataset("WNJXYK/TTA-ImageNet-C",
223
+ name="gaussian_noise",
224
+ split="severity_5",
225
+ revision="v1.0")
226
+ ```
227
+
228
+ ## Provenance
229
+
230
+ This mirror was built by `scripts/publish_imagenetc.py` in the
231
+ TTA-Evaluation-Harness repo. JPEG bytes are copied 1:1 from the upstream
232
+ files - no re-encoding, pixel-for-pixel identical to Hendrycks's release.