bstds commited on
Commit
b82a88a
·
1 Parent(s): 0fb0281

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +310 -0
README.md ADDED
@@ -0,0 +1,310 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Amazon Berkeley Objects (c) by Amazon.com
2
+
3
+ ## License
4
+
5
+ This work is licensed under the Creative Commons Attribution-NonCommercial 4.0
6
+ International Public License. To obtain a copy of the full license, see
7
+ `LICENSE-CC-BY-NC-4.0.txt`, visit
8
+ [CreativeCommons.org](https://creativecommons.org/licenses/by-nc/4.0/)
9
+ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.
10
+
11
+ Under the following terms:
12
+
13
+ * Attribution — You must give appropriate credit, provide a link to the
14
+ license, and indicate if changes were made. You may do so in any reasonable
15
+ manner, but not in any way that suggests the licensor endorses you or your
16
+ use.
17
+
18
+ * NonCommercial — You may not use the material for commercial purposes.
19
+
20
+ * No additional restrictions — You may not apply legal terms or technological
21
+ measures that legally restrict others from doing anything the license
22
+ permits.
23
+
24
+ ## Attribution
25
+
26
+ Credit for the data, including all images and 3d models, must be given to:
27
+
28
+ > Amazon.com
29
+
30
+ Credit for building the dataset, archives and benchmark sets must be given to:
31
+
32
+ > Matthieu Guillaumin (Amazon.com), Thomas Dideriksen (Amazon.com),
33
+ > Kenan Deng (Amazon.com), Himanshu Arora (Amazon.com),
34
+ > Jasmine Collins (UC Berkeley) and Jitendra Malik (UC Berkeley)
35
+
36
+ ## Description
37
+
38
+ Amazon Berkeley Objects is a collection of 147,702 product listings with
39
+ multilingual metadata and 398,212 unique catalog images. 8,222 listings come
40
+ with turntable photography (also referred as *spin* or *360º-View* images), as
41
+ sequences of 24 or 72 images, for a total of 586,584 images in 8,209 unique
42
+ sequences. For 7,953 products, the collection also provides high-quality 3d
43
+ models, as glTF 2.0 files.
44
+
45
+ The collection is made of the following files:
46
+
47
+ * `README.md` - The present file.
48
+
49
+ * `LICENSE-CC-BY-NC-4.0.txt` - The License file. You must read, agree and
50
+ comply to the License before using the Amazon Berkeley Objects data.
51
+
52
+ * `listings/metadata/listings_<i>.json.gz` - Product description and metadata.
53
+ Each of the 16 files is encoded with UTF-8 and gzip-compressed. Each line of
54
+ the decompressed files corresponds to one product as a JSON object (see
55
+ http://ndjson.org/ or https://jsonlines.org/ ). Each product JSON object
56
+ (a.k.a dictionary) has any number of the following keys:
57
+
58
+ - `brand`
59
+ - Content: Brand name
60
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
61
+ - `bullet_point`
62
+ - Content: Important features of the products
63
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
64
+ - `color`
65
+ - Content: Color of the product as text
66
+ - Format: `[{"language_tag": <str>, "standardized_values": [<str>],
67
+ "value": <str>}, ...]`
68
+ - `color_code`
69
+ - Content: Color of the product as HTML color code
70
+ - Format: `[<str>, ...]`
71
+ - `country`
72
+ - Content: Country of the marketplace, as an
73
+ [ISO 3166-1 alpha 2](https://en.wikipedia.org/wiki/ISO_3166-1_alpha-2)
74
+ code
75
+ - Format: `<str>`
76
+ - `domain_name`
77
+ - Content: Domain name of the marketplace where the product is found.
78
+ A product listing in this collection is uniquely identified by
79
+ (`item_id`, `domain_name`)
80
+ - Format: `<str>`
81
+ - `fabric_type`
82
+ - Content: Description of product fabric
83
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
84
+ - `finish_type`
85
+ - Content: Description of product finish
86
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
87
+ - `item_dimensions`
88
+ - Content: Dimensions of the product (height, width, length)
89
+ - Format: `{"height": {"normalized_value": {"unit": <str>, "value":
90
+ <float>}, "unit": <str>, "value": <float>}, "length":
91
+ {"normalized_value": {"unit": <str>, "value": <float>}, "unit": <str>,
92
+ "value": <float>}, "width": {"normalized_value": {"unit": <str>,
93
+ "value": <float>}, "unit": <str>, "value": <float>}}}`
94
+ - `item_id`
95
+ - Content: The product reference id. A product listing in this
96
+ collection is uniquely identified by (`item_id`, `domain_name`).
97
+ A corresponding product page may exist at
98
+ `https://www.<domain_name>/dp/<item_id>`
99
+ - Format: `<str>`
100
+ - `item_keywords`
101
+ - Content: Keywords for the product
102
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
103
+ - `item_name`
104
+ - Content: The product name
105
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
106
+ - `item_shape`
107
+ - Content: Description of the product shape
108
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
109
+ - `item_weight`
110
+ - Content: The product weight
111
+ - Format: `[{"normalized_value": {"unit": <str>, "value": <float>},
112
+ "unit": <str>, "value": <float>}, ...]`
113
+ - `main_image_id`
114
+ - Content: The main product image, provided as an `image_id`. See the
115
+ descripton of `images/metadata/images.csv.gz` below
116
+ - Format: `<str>`
117
+ - `marketplace`
118
+ - Content: Retail website name (Amazon, AmazonFresh, AmazonGo, ...)
119
+ - Format: `<str>`
120
+ - `material`
121
+ - Content: Description of the product material
122
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
123
+ - `model_name`
124
+ - Content: Model name
125
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
126
+ - `model_number`
127
+ - Content: Model number
128
+ - Format: `[{ "language_tag": <str>, "value": <str> }, ...]`
129
+ - `model_year`
130
+ - Content: Model year
131
+ - Format: `[{ "language_tag": <str>, "value": <int> }, ...]`
132
+ - `node`
133
+ - Content: Location of the product in the category tree. A node page
134
+ may exist at `https://www.<domain_name>/b/?node=<node_id>` for
135
+ browsing
136
+ - Format: `[{ "node_id": <int>, "path": <str>}, ...]`
137
+ - `other_image_id`
138
+ - Content: Other available images for the product, provided as
139
+ `image_id`. See the description of `images/metadata/images.csv.gz`
140
+ below
141
+ - Format: `[<str>, ...]`
142
+ - `pattern`
143
+ - Content: Product pattern
144
+ - Format: `[{ "language_tag": <str>, "value": <int> }, ...]`
145
+ - `product_description`
146
+ - Content: Product description as HTML
147
+ - Format: `[{ "language_tag": <str>, "value": <int> }, ...]`
148
+ - `product_type`
149
+ - Content: Product type (category)
150
+ - Format: `<str>`
151
+ - `spin_id`
152
+ - Content: Reference to the 360º View image sequence. See the
153
+ description of `spins/metadata/spins.csv.gz` below
154
+ - Format: `<str>`
155
+ - `style`
156
+ - Content: Style of the product
157
+ - Format: `[{ "language_tag": <str>, "value": <int> }, ...]`
158
+ - `3dmodel_id`
159
+ - Content: Reference to the 3d model of the product. See the description
160
+ of `3dmodels/metadata/3models.csv.gz`
161
+ - Format: `<str>`
162
+
163
+ * `images/metadata/images.csv.gz` - Image metadata. This file is a
164
+ gzip-compressed comma-separated value (CSV) file with the following
165
+ columns: `image_id`, `height`, `width`, and `path`.
166
+ - `image_id` (string): this id uniquely refers to a product image. This id
167
+ can be used to retrieve the image data from Amazon's Content Delivery
168
+ Network (CDN) using the template:
169
+ `https://m.media-amazon.com/image/I/<image_id>.<extension>` [^1],
170
+ where `<extension>` is composed of the characters following the dot in the
171
+ `path` field. Any value occurring in the `main_image` and `other_images`
172
+ attributes of product metadata is an `image_id` present in this file.
173
+ - `height` (int) and `width` (int): respectively, the height and width of
174
+ the original image.
175
+ - `path`: the location of the image file relative to the `images/original/`
176
+ or `images/small/` directories. A path is composed of lowercase hex
177
+ characters (`0-9a-f`) that also uniquely identifies images. The first two
178
+ characters are used to build a file hierarchy and reduce the number of
179
+ images in a single directory. The extension is `jpg` except for few `png`
180
+ files.
181
+
182
+ Below are are first 10 lines of `images/metadata/images.csv.gz`:
183
+ ```
184
+ image_id,height,width,path
185
+ 010-mllS7JL,106,106,14/14fe8812.jpg
186
+ 01dkn0Gyx0L,122,122,da/daab0cad.jpg
187
+ 01sUPg0387L,111,111,d2/d2daaae9.jpg
188
+ 1168jc-5r1L,186,186,3a/3a4e88e6.jpg
189
+ 11RUV5Fs65L,30,500,d9/d91ab9cf.jpg
190
+ 11X4pFHqYOL,35,500,20/20098c4d.jpg
191
+ 11Y+Xpt1lfL,103,196,99/9987a1c8.jpg
192
+ 11rL64ZLPYL,64,500,89/89a2ff4d.jpg
193
+ 11xjmNF5TAL,117,88,ee/ee239f0f.jpg
194
+ ```
195
+
196
+ * `images/original/<path>` - Original image data. This directory contains the
197
+ original high-resolution version of the images. See
198
+ `images/metadata/images.csv.gz` for details of image naming.
199
+
200
+ * `images/small/<path>` - Downscaled image data. This directory contains the
201
+ version of the images, where they have been downscaled such that their
202
+ largest axis (height or width) is a maximum of 256 pixels. See
203
+ `images/metadata/images.csv.gz` for details of image naming.
204
+
205
+ * `spins/metadata/spins.csv.gz` - Spin / 360º-View image metadata. This file
206
+ is a gzip-compressed comma-separated value (CSV) file with the following
207
+ fields: `spin_id`, `azimuth`, `image_id`, `height`, `width`, and `path`.
208
+ - `spin_id`: a unique identifier for the image sequence.
209
+ - `azimuth`: an integer between 0 and 71, representing the index in the spin
210
+ sequence and the azimuth of the camera (in steps of 5º).
211
+ - `image_id`: this id uniquely refers to an image. This is can be used
212
+ to retrieve the image data using the template:
213
+ `https://m.media-amazon.com/image/I/<image_id>.jpg` [^1].
214
+ - `height` and `width`: respectively, the height and width of the image.
215
+ - `path`: the location of the image file relative to the `spins/original/`
216
+ directory. The extension is `jpg` for all the images. The `path` is build
217
+ from the `spin_id` and `azimuth` using the template
218
+ `<spin_id>_<azimuth:02d>.jpg` and the first two characters of `spin_id`
219
+ are used to build a file hierarchy and reduce the number of files in a
220
+ single directory.
221
+
222
+ Below are are first 10 lines of `spins/metadata/spins.csv.gz`:
223
+ ```
224
+ spin_id,azimuth,image_id,height,width,path
225
+ 61c91265,0,41wqHws7a6L,248,1075,61/61c91265/61c91265_00.jpg
226
+ 61c91265,1,41++eZZHP9L,248,1075,61/61c91265/61c91265_01.jpg
227
+ 61c91265,2,41YF86LhGDL,248,1075,61/61c91265/61c91265_02.jpg
228
+ 61c91265,3,41I5Zz-kbAL,248,1075,61/61c91265/61c91265_03.jpg
229
+ 61c91265,4,41lAQM2Ys5L,248,1075,61/61c91265/61c91265_04.jpg
230
+ 61c91265,5,41OJT+p8JgL,248,1075,61/61c91265/61c91265_05.jpg
231
+ 61c91265,6,412kYqOnqHL,248,1075,61/61c91265/61c91265_06.jpg
232
+ 61c91265,7,41rgUZ0NuFL,248,1075,61/61c91265/61c91265_07.jpg
233
+ 61c91265,8,41PJ4ks-cWL,248,1075,61/61c91265/61c91265_08.jpg
234
+ ```
235
+
236
+ * `spins/original/<path>` - Spin / 360º-View image files. Each file
237
+ corresponds to one row in `spins/metadata/spins.csv.gz`, named by the value
238
+ of the `path` column.
239
+
240
+ * `3dmodels/metadata/3dmodels.csv.gz` - 3d model metadata. This file is a
241
+ gzip-compressed comma-separated value (CSV) file with the following fields:
242
+ `3dmodel_id`, `path`, `meshes`, `materials`, `textures`, `images`,
243
+ `image_height_max`, `image_height_min`, `image_width_max`,
244
+ `image_width_min`, `vertices`, `faces`, `extent_x`, `extent_y`, `extent_z`.
245
+
246
+ - `3dmodel_id`: Reference for the 3d model, as provided in the `3dmodel_id`
247
+ field of the listings metadata
248
+ - `path`: Location of the 3d model, relative to `3dmodels/original/`
249
+ - `meshes`: Number of meshes in the geometry
250
+ - `materials`: Number of materials in the 3d model
251
+ - `textures`: Number of textures in the 3d model
252
+ - `images`: Number of image resources in the 3d model
253
+ - `image_{heigh,width}_{min,max}`: Minimal and maximal dimensions of the
254
+ image resources in the 3d model
255
+ - `vertices`: Number of vertices in the geometry
256
+ - `faces`: Number of faces in the geometry
257
+ - `extent_{x,y,z}`: Extent of the geometry in each dimension
258
+
259
+ Below are are first 10 lines of `3dmodels/metadata/3dmodels.csv.gz`:
260
+ ```
261
+ 3dmodel_id,path,meshes,materials,textures,images,image_height_max,image_height_min,image_width_max,image_width_min,vertices,faces,extent_x,extent_y,extent_z
262
+ B01N2PLWIL,L/B01N2PLWIL.glb,1,1,3,3,4096,4096,4096,4096,10990,14380,0.571499943733216,0.11684001048680606,0.07111982014177629
263
+ B075QFCHM9,9/B075QFCHM9.glb,1,1,3,3,2048,2048,2048,2048,11973,19568,1.840071976184845,1.0669103860855103,2.3675915002822876
264
+ B07H469871,1/B07H469871.glb,1,1,3,3,4096,4096,4096,4096,1602,1950,1.1113524436950684,1.3880813121795654,0.39794909954071045
265
+ B07H8V49M2,2/B07H8V49M2.glb,1,1,3,3,4096,4096,4096,4096,3760,5710,1.4998703368605968,2.11988401412964,0.5897879977087402
266
+ B07DBHPK4G,G/B07DBHPK4G.glb,1,1,3,3,4096,4096,4096,4096,13704,22736,0.37921489775180817,1.6228150129318237,0.37921497225761414
267
+ B0842LM2DN,N/B0842LM2DN.glb,1,1,3,3,4096,4096,4096,4096,4078,7584,0.22779017686843872,0.2348586767911911,0.22779015451669693
268
+ B07HK6B4D7,7/B07HK6B4D7.glb,1,1,4,4,2048,2048,2048,2048,12221,19268,0.1887289583683014,0.6650936603546143,0.421692430973053
269
+ B07B4FZN9H,H/B07B4FZN9H.glb,1,1,3,3,2048,2048,2048,2048,13595,22644,3.3838289976119995,0.9963648915290833,2.048073887825012
270
+ B07B4Z9BS4,4/B07B4Z9BS4.glb,1,1,4,4,2048,2048,2048,2048,9259,16178,0.2793540060520172,0.2693932056427002,0.2793540358543396
271
+ ```
272
+
273
+ * `3dmodels/original/<path>` - 3d model files. The 3d models are provided in
274
+ the [glTF-2.0 format](
275
+ https://github.com/KhronosGroup/glTF/blob/master/specification/2.0/README.md
276
+ ) (GLB/binary representation). All models adhere to the following
277
+ conventions:
278
+ 1. Positive `Y` direction is up.
279
+ 2. Positive `Z` direction is pointing to the *natural* front-side of the
280
+ product, wherever applicable.
281
+ 3. Products that are designed to stand on a surface (i.e. a floor) are
282
+ centered on the origin, but translated up (towards positive `Y`) such
283
+ that they *stand* on the `Y=0` plane.
284
+ 4. Products that are designed to hang from a surface (i.e. a ceiling) are
285
+ centered on the origin, but translated down (towards negative `Y`) such
286
+ that they *hang* from the `Y=0` plane.
287
+ 5. Products that are designed to hang on a wall are centered on the origin,
288
+ but translated forward (towards positive `Z`) such that their backside
289
+ aligns with the `Z=0` plane.
290
+
291
+ * `archives/abo-listings.tar` - Contains all the files in `listings/` as a
292
+ tar archive.
293
+
294
+ * `archives/abo-images-original.tar` - Contains the metadata and original
295
+ images from `images/original/` as a tar archive.
296
+
297
+ * `archives/abo-images-small.tar` - Contains the metadata and downscaled
298
+ images from `images/small/` as a tar archive.
299
+
300
+ * `archives/abo-spins.tar` - Contains the metadata and images from `spins/`
301
+ as a tar archive.
302
+
303
+ * `archives/abo-3dmodels.tar` - Contains the metadata and images from `3dmodels/`
304
+ as a tar archive.
305
+
306
+ ## Footnotes
307
+
308
+ [^1]: Importantly, there is no guarantee that those URLs will remain unchanged
309
+ and available on the long term, we thus recommend using the images provided in this
310
+ archive instead.