README.md

metadata

license: cc-by-4.0
pretty_name: Amazon Berkeley Objects
tags:
  - computer-vision
  - image-retrieval
  - product-catalog
  - hierarchy
  - 3d
configs:
  - config_name: listings
    data_files:
      - split: train
        path: data/listings/*.parquet
  - config_name: images
    data_files:
      - split: train
        path: data/images/*.parquet
  - config_name: spins
    data_files:
      - split: train
        path: data/spins/*.parquet
  - config_name: 3dmodels
    data_files:
      - split: train
        path: data/3dmodels/*.parquet

Amazon Berkeley Objects

This is a Hugging Face metadata mirror of the Amazon Berkeley Objects dataset for reproducible research and HyperView demos. The original dataset is provided by Amazon.com and UC Berkeley.

This mirror stores metadata tables and official S3 asset URLs. It does not duplicate catalog images, turntable images, or 3D models as binary files.

Load

from datasets import load_dataset

listings = load_dataset("hyper3labs/amazon-berkeley-objects", "listings", split="train")
images = load_dataset("hyper3labs/amazon-berkeley-objects", "images", split="train")
spins = load_dataset("hyper3labs/amazon-berkeley-objects", "spins", split="train")
models = load_dataset("hyper3labs/amazon-berkeley-objects", "3dmodels", split="train")

Tables

• listings: all product listings with normalized hierarchy columns and raw_listing_json preserving the original listing object.
• images: full image metadata with official small and original image URLs.
• spins: full spin / 360-degree view metadata with official image URLs.
• 3dmodels: full 3D model metadata with official GLB URLs.

Added convenience columns

The listings table adds:

• title
• brand
• color
• style
• product_type
• product_type_readable
• node_paths
• department
• hierarchy_path
• main_image_path
• main_image_url

License

Amazon Berkeley Objects is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).

Users must read and comply with the original ABO license and attribution requirements before using the data.

Official dataset page: https://amazon-berkeley-objects.s3.amazonaws.com/index.html

AWS Open Data Registry: https://registry.opendata.aws/amazon-berkeley-objects/

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 Guillaumin, Thomas Dideriksen, Kenan Deng, Himanshu Arora, Jasmine Collins, and Jitendra Malik, with the complete author list in the ABO paper and official documentation.