---
license: cc-by-4.0
pretty_name: Amazon Berkeley Objects (ABO)
source_datasets:
  - original
task_categories:
  - image-classification
  - image-to-3d
tags:
  - amazon
  - products
  - 3d
  - images
  - ecommerce
  - abo
configs:
  - config_name: mvr
    data_files:
      - split: train
        path: mvr/train-*.parquet
  - config_name: listings
    data_files:
      - split: train
        path: listings/train-*.parquet
  - config_name: part_labels
    data_files:
      - split: train
        path: part_labels/train-*.parquet
  - config_name: images_small
    data_files:
      - split: train
        path: images_small/train-*.parquet
  - config_name: spins
    data_files:
      - split: train
        path: spins/train-*.parquet
  - config_name: images_original
    data_files:
      - split: train
        path: images_original/train-*.parquet
  - config_name: models_3d
    data_files:
      - split: train
        path: models_3d/train-*.parquet
  - config_name: benchmark_material
    data_files:
      - split: train
        path: benchmark_material/train-*.parquet
  - config_name: objects
    default: true
    data_files:
      - split: train
        path: objects/train-*.parquet
---

# Amazon Berkeley Objects (ABO)

A Hugging Face packaging of the **Amazon Berkeley Objects (ABO)** dataset. The
data content is the official CC BY 4.0 release from
<https://amazon-berkeley-objects.s3.amazonaws.com/index.html>. This mirror
changes only the packaging: files are grouped into typed Parquet shards, and
**every original media file is preserved byte-for-byte and never transcoded**.
Images use the `datasets` `Image()` feature, 3D product models use the native
`Mesh()` feature (original GLB bytes), and part-label meshes are kept in their
original `.obj` format — so the Hugging Face Dataset Viewer and `datasets`
streaming APIs can consume them.

## Configs

| Config | Row unit | Main fields |
| --- | --- | --- |
| `objects` (default) | One product | Listing summary, foreign keys, modality flags |
| `listings` | One product listing | Typed fields plus lossless `raw_listing_json` |
| `images_small` | One catalog image | 256 px `Image()`, `image_id`, dimensions, path |
| `images_original` | One catalog image | Original-resolution `Image()`, dimensions, path |
| `spins` | One turntable frame | `Image()`, `spin_id`, `image_id`, `azimuth` |
| `models_3d` | One 3D product | Native `Mesh()` GLB and geometry/material stats |
| `part_labels` | One part file | Original `.obj` part-mesh bytes (verbatim) or `relation.json`; keyed by `item_id`/`part_id` |
| `benchmark_material` | One model-viewpoint file | Render/material `Image()` or EXR `blob`, `kind` |
| `mvr` | One retrieval example | Official CVPR22 multi-view-retrieval split columns |

`models_3d` uses the native `Mesh()` feature (requires a recent `datasets`
release with mesh support). Part-label meshes are stored in their original
`.obj` format under `part_labels`, alongside each item's `relation.json`. Note:
the Hugging Face **Dataset Viewer does not yet render the `Mesh` feature**; 3D
models can still be previewed via the Hub's per-file viewer and loaded through
`datasets`.

## Licensing

This work is licensed under the **Creative Commons Attribution 4.0 International
Public License (CC BY 4.0)**. A verbatim copy is provided in
[`LICENSE-CC-BY-4.0.txt`](./LICENSE-CC-BY-4.0.txt); see also
<https://creativecommons.org/licenses/by/4.0/>.

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.

**Changes made:** the original release was repackaged into typed Parquet shards
(`Image()`/`Mesh()` features) for the Hugging Face Dataset Viewer and
`datasets`. No underlying data content was altered.

> Note: the source S3 bucket root also contains a `LICENSE-CC-BY-NC-4.0.txt`
> file, but the official ABO download page licenses the released archives under
> **CC BY 4.0**, which is the license applied and mirrored here.

## 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 (Amazon.com), Thomas Dideriksen (Amazon.com),
> Kenan Deng (Amazon.com), Himanshu Arora (Amazon.com), Arnab Dhua (Amazon.com),
> Xi (Brian) Zhang (Amazon.com), Tomas Yago-Vicente (Amazon.com),
> Jasmine Collins (UC Berkeley), Shubham Goel (UC Berkeley),
> Jitendra Malik (UC Berkeley).

No endorsement by Amazon.com or UC Berkeley of this mirror is claimed or implied.

## Citation

```bibtex
@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}
}
```