---
license: cc-by-4.0
pretty_name: Deep1B (mirror)
size_categories:
  - 1B<n<10B
task_categories:
  - feature-extraction
tags:
  - ann-benchmark
  - vector-search
  - billion-scale
---

# Deep1B — mirror

Mirror of the **Deep1B** billion-scale ANN benchmark (1,000,000,000 × 96-dim
`float32` deep image descriptors) and its 10,000-query public test set,
originally published by Yandex Research and hosted at
`https://storage.yandexcloud.net/yandex-research/ann-datasets/DEEP/`.

This is a verbatim byte-for-byte copy intended for ANN-benchmark
reproducibility. **Original authors / license**: Yandex Research; please
cite their work (Babenko & Lempitsky, "Efficient Indexing of Billion-Scale
Datasets of Deep Descriptors", CVPR 2016) and respect their terms of use.

## Files

| file | size | format |
|---|---:|---|
| `base.1B.fbin` | 384,000,000,008 B | `fbin`: int32 `n=1,000,000,000`, int32 `d=96`, then `n*d` float32 |
| `query.public.10K.fbin` | 3,840,008 B | `fbin`: `n=10,000`, `d=96`, then `n*d` float32 |

Mirrored on 2026-05-22.