Amazon Berkeley Objects (ABO) Dataset

A CC BY 4.0-licensed dataset of Amazon products with metadata, catalog images, and 3D models.

Product Metadata

{
  "item_id": "B075X4QMX3",
  "domain_name": "amazon.com",
  "item_name": [
    {
      "language_tag": "en_US",
      "value": "Stone & Beam Westport Modern Nailhead Upholstered Sofa, 87\"W, Linen"
    }
    {
      "language_tag": "zh_CN",
      "value": "亚马逊品牌 - Stone & Beam Westport 现代钉头座椅系列, 亚麻, Neutral"
    },
    {
      "language_tag": "ko_KR",
      "value": "아마존 브랜드 - 스톤 앤 빔 Westport 모던 네일헤드 시트 컬렉션"
    },
    {
      "language_tag": "he_IL",
      "value": "מותג אמזון – Stone & Beam Westport אוסף מושבים מודרני"
    },
    {
      "language_tag": "de_DE",
      "value": "Amazon Marke - Stone & Beam Westport Modern Nailhead Sitzgruppe Kollektion, Leinen, neutral"
    },
    {
      "language_tag": "es_US",
      "value": "Marca Amazon – Stone & Beam Westport colección de asientos modernos para clavos, lino, Neutral"
    },
  ],
  ...
}

Metadata includes multilingual title, brand, model, year, product type, color, description, dimensions, weight, material, pattern, and style.

Catalog Images

For the 147,702 products, we provide 398,212 unique catalog images in high resolution.

360º Images

For more than 8,200 products, the dataset includes a sequence of 72 images, capturing the product every 5º in azimuth, for a total of 586,584 images.

3D Models

The dataset contains high-quality 3D models with 4K texture maps for physically based rendering for more than 7'900 products. The models are provided in the standard glTF 2.0 format.

Renderings

For products with 3D models, we provide rendered images for 91 viewpoints on the upper icosphere, with varying azimuth and elevation.

Environments

For each model and viewpoint, 3 different environment maps are used to provide renderings with varied lighting, for a total of 2.1 million images.

Geometry

For each rendering, we provide the camera parameters, the object segmentation mask as well as dense normals and depth maps.

Materials

As well as texture maps of SVBRDF properties: base color, metallic and roughness.

Explore

Explore the 147,702 products in ABO by specifying keywords of product names, product type and choosing to show only products with 360º view images and/or 3D models.

Instructions

Activate the exploration tool by loading the metadata (68 Mb)
Use the filter form to narrow down on products of interest
Click on product thumbnails to toggle the visualisation of product metadata (partial), images, 360º images and 3D models when available.

Filters

Filter products by name (any language)

Filter products by type

Only products with 360º images

Only products with 3D models

Download

This work is licensed under the Creative Commons Attribution 4.0 International Public License (CC BY 4.0). To obtain a copy of this license, see LICENSE-CC-BY-4.0.txt in the archive, visit CreativeCommons.org or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

Under the following terms:

Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits.

The following archives are available for download:

LICENSE-CC-BY-4.0.txt — Copy of the CC BY 4.0 license
README.md — Description of the dataset
abo-listings.tar — Product listings and metadata (83 Mb)
abo-images-original.tar — Original catalog images and metadata (110 Gb)
abo-images-small.tar — Downscaled (max 256 pixels) catalog images and metadata (3 Gb)
abo-spins.tar — 360º-view images and metadata (40 Gb)
abo-3dmodels.tar — 3D models (154 Gb)
abo-mvr.csv.xz — Dataset splits for the CVPR22 Multi-View Retrieval experiments (4 Mb)
abo-benchmark-material.tar — Rendered 3D models for the CVPR22 Material Prediction experiments (271 Gb)
abo-part-labels.tar — 3D part labels for the 3D Vision and Modeling Challenges in eCommerce ICCV 2023 Workshop Challenge (2 Gb)

Attribution

Credit for the data, including all images and 3D models, must be given to:

Amazon.com

Credit for building the dataset, archives and benchmark sets must be given to:

Matthieu GuillauminAmazon.com

Thomas DideriksenAmazon.com

Kenan DengAmazon.com

Himanshu AroraAmazon.com

Arnab DhuaAmazon.com

Xi (Brian) ZhangAmazon.com

Tomas Yago-VicenteAmazon.com

Jasmine CollinsUC Berkeley

Shubham GoelUC Berkeley

Jitendra MalikUC Berkeley

If you use this dataset for publication, we kindly ask you to attribute credit by citing our CVPR 2022 paper, or the version on arxiv.org, providing detailed description of the dataset and benchmarks:

@article{collins2022abo,
  title={ABO: Dataset and Benchmarks for Real-World 3D Object Understanding},
  author={Collins, Jasmine and Goel, Shubham and Deng, Kenan and Luthra, Achleshwar and
          Xu, Leon and Gundogdu, Erhan and Zhang, Xi and Yago Vicente, Tomas F and
          Dideriksen, Thomas and Arora, Himanshu and Guillaumin, Matthieu and
          Malik, Jitendra},
  journal={CVPR},
  year={2022}
}

Publications and links

Below is the list of publications known to use ABO:

ABO: Dataset and Benchmarks for Real-World 3D Object Understanding — The reference paper for the ABO dataset (CVPR 2022).

Externally contributed data and code for ABO can be found at those locations:

Additional ABO data and annotations — Repository with WordNet/ShapeNet class labels for 3D models.
ABO 3D Renderings — The rendered ABO images and camera poses for 3D reconstruction benchmarking in the CVPR 2022 paper.

Workshops and challenges using ABO data:

3D Vision and Modeling Challenges in eCommerce — Workshop and challenge on 3D part labeling, in conjunction with ICCV 2023

Cloud Usage (AWS)

The ABO dataset is directly available on Amazon S3 at s3://amazon-berkeley-objects/, with the same structure as in the archives (see README).

In particular, the metadata is loadable in Amazon Athena, using the following SQL table creation statements:

CREATE EXTERNAL TABLE IF NOT EXISTS `default`.`abo_listings` (
  `brand` array < struct < language_tag:string, value:string >  >,
  `bullet_point` array < struct < language_tag:string, value:string >  >,
  `color` array < struct < language_tag:string, value:string >  >,
  `color_code` array < string >,
  `country` string,
  `domain_name` string,
  `fabric_type` array < struct < language_tag:string, value:string >  >,
  `finish_type` array < struct < language_tag:string, value:string >  >,
  `item_dimensions` struct < height:struct < normalized_value:struct < unit:string, value:float >,  value:float, unit:string >,  length:struct < normalized_value:struct < unit:string, value:float >,  value:float, unit:string >,  width:struct < normalized_value:struct < unit:string, value:float >,  value:float, unit:string >  >,
  `item_id` string,
  `item_keywords` array < struct < language_tag:string, value:string >  >,
  `item_name` array < struct < language_tag:string, value:string >  >,
  `item_shape` array < struct < language_tag:string, value:string >  >,
  `item_weight` array < struct < normalized_value:struct < unit:string, value:float >,  value:float, unit:string >  >,
  `main_image_id` string,
  `marketplace` string,
  `material` array < struct < language_tag:string, value:string >  >,
  `model_name` array < struct < language_tag:string, value:string >  >,
  `model_number` array < struct < language_tag:string, value:string >  >,
  `model_year` array < struct < language_tag:string, value:string >  >,
  `node` array < struct < node_id:bigint, path:string >  >,
  `other_image_id` array < string >,
  `pattern` array < struct < language_tag:string, value:string >  >,
  `product_description` array < struct < language_tag:string, value:string >  >,
  `product_type` array < struct < value:string >  >,
  `spin_id` string,
  `style` array < struct < language_tag:string, value:string >  >,
  `3dmodel_id` string
)
ROW FORMAT SERDE
  'org.openx.data.jsonserde.JsonSerDe'
WITH SERDEPROPERTIES (
  'serialization.format' = '1'
)
LOCATION
  's3://amazon-berkeley-objects/listings/metadata/'
TBLPROPERTIES (
  'has_encrypted_data'='false'
)

Note: you can change the statement database (`default`) and table name (`abo_listings`) with values of your choice.

CREATE EXTERNAL TABLE IF NOT EXISTS `default`.`abo_images`(
  `image_id` string,
  `height` bigint,
  `width` bigint,
  `path` string
)
ROW FORMAT SERDE
  'org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe'
WITH SERDEPROPERTIES (
  'serialization.format' = ',',
  'field.delim' = ',',
  'skip.header.line.count'='1'
)
LOCATION
  's3://amazon-berkeley-objects/images/metadata/'
TBLPROPERTIES (
  'has_encrypted_data'='false'
)

Note: Note: you can change the statement database (`default`) and table name (`abo_images`) with values of your choice.

CREATE EXTERNAL TABLE IF NOT EXISTS `default`.`abo_spins`(
  `spin_id` string,
  `azimuth` bigint,
  `image_id` string,
  `height` bigint,
  `width` bigint,
  `path` string
)
ROW FORMAT SERDE
  'org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe'
WITH SERDEPROPERTIES (
  'serialization.format' = ',',
  'field.delim' = ',',
  'skip.header.line.count'='1'
)
LOCATION
  's3://amazon-berkeley-objects/spins/metadata/'
TBLPROPERTIES (
  'has_encrypted_data'='false'
)

Note: Note: you can change the statement database (`default`) and table name (`abo_spins`) with values of your choice.

CREATE EXTERNAL TABLE IF NOT EXISTS `default`.`abo_3dmodels`(
  `3dmodel_id` string,
  `path` string,
  `meshes` bigint,
  `materials` bigint,
  `textures` bigint,
  `images` bigint,
  `image_height_max` bigint,
  `image_height_min` bigint,
  `image_width_max` bigint,
  `image_width_min` bigint,
  `vertices` bigint,
  `faces` bigint,
  `extent_x` float,
  `extent_y` float,
  `extent_z` float
  )
ROW FORMAT SERDE
  'org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe'
WITH SERDEPROPERTIES (
  'serialization.format' = ',',
  'field.delim' = ',',
  'skip.header.line.count'='1'
)
LOCATION
  's3://amazon-berkeley-objects/3dmodels/metadata/'
TBLPROPERTIES (
  'has_encrypted_data'='false'
)

Note: Note: you can change the statement database (`default`) and table name (`abo_3dmodels`) with values of your choice.

Acknowledgements and Credits

This webpage is built with:

^†: including GLTFLoader, RGBELoader and OrbitControls.
^*: glb-viewer was modified to add support for OrbitControls, lighting from environment map and deferred rendering.