README.md

---
dataset_info:
- config_name: candidates
  features:
  - name: did
    dtype: string
  - name: text
    list: string
  - name: media_id
    dtype: string
  splits:
  - name: train
    num_bytes: 931441504
    num_examples: 7119841
  download_size: 499532524
  dataset_size: 931441504
- config_name: instructions
  features:
  - name: task_id
    dtype: int32
  - name: dataset_id
    dtype: int32
  - name: prompts
    dtype: string
  splits:
  - name: train
    num_bytes: 28058
    num_examples: 58
  download_size: 11004
  dataset_size: 28058
- config_name: media_shard_000000
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 122849524199
    num_examples: 200000
  download_size: 122853793676
  dataset_size: 122849524199
- config_name: media_shard_000001
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117985237498
    num_examples: 200000
  download_size: 117988606222
  dataset_size: 117985237498
- config_name: media_shard_000002
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117669748066
    num_examples: 200000
  download_size: 117673236807
  dataset_size: 117669748066
- config_name: media_shard_000003
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117765874023
    num_examples: 200000
  download_size: 117769664261
  dataset_size: 117765874023
- config_name: media_shard_000004
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117586015352
    num_examples: 200000
  download_size: 117590569299
  dataset_size: 117586015352
- config_name: media_shard_000005
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 118252475623
    num_examples: 200000
  download_size: 118255300760
  dataset_size: 118252475623
- config_name: media_shard_000006
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117639713891
    num_examples: 200000
  download_size: 117643189579
  dataset_size: 117639713891
- config_name: media_shard_000007
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117427707334
    num_examples: 200000
  download_size: 117431098509
  dataset_size: 117427707334
- config_name: media_shard_000008
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 118041395077
    num_examples: 200000
  download_size: 118046142117
  dataset_size: 118041395077
- config_name: media_shard_000009
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117724310015
    num_examples: 200000
  download_size: 117728340239
  dataset_size: 117724310015
- config_name: media_shard_000010
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117390521616
    num_examples: 200000
  download_size: 117395658377
  dataset_size: 117390521616
- config_name: media_shard_000011
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117439367419
    num_examples: 200000
  download_size: 117444566123
  dataset_size: 117439367419
- config_name: media_shard_000012
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 118444042857
    num_examples: 200000
  download_size: 118449572108
  dataset_size: 118444042857
- config_name: media_shard_000013
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117590368881
    num_examples: 200000
  download_size: 117594645617
  dataset_size: 117590368881
- config_name: media_shard_000014
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117985815582
    num_examples: 200000
  download_size: 117990857457
  dataset_size: 117985815582
- config_name: media_shard_000015
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117263260177
    num_examples: 200000
  download_size: 117267034089
  dataset_size: 117263260177
- config_name: media_shard_000016
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 117574792839
    num_examples: 200000
  download_size: 117579302317
  dataset_size: 117574792839
- config_name: media_shard_000017
  features:
  - name: media_id
    dtype: string
  - name: image
    dtype: image
  - name: video
    list: image
  - name: audio
    dtype: binary
  - name: has_audio
    dtype: bool
  splits:
  - name: train
    num_bytes: 81693044328
    num_examples: 138323
  download_size: 81696151897
  dataset_size: 81693044328
- config_name: queries
  features:
  - name: id
    dtype: string
  - name: task_id
    dtype: int32
  - name: text
    list: string
  - name: media_id
    dtype: string
  - name: candidates
    list: string
  splits:
  - name: train
    num_bytes: 834160494
    num_examples: 6405109
  download_size: 366840294
  dataset_size: 834160494
configs:
- config_name: candidates
  data_files:
  - split: train
    path: candidates/train-*
- config_name: instructions
  data_files:
  - split: train
    path: instructions/train-*
- config_name: media_shard_000000
  data_files:
  - split: train
    path: media_shard_000000/train-*
- config_name: media_shard_000001
  data_files:
  - split: train
    path: media_shard_000001/train-*
- config_name: media_shard_000002
  data_files:
  - split: train
    path: media_shard_000002/train-*
- config_name: media_shard_000003
  data_files:
  - split: train
    path: media_shard_000003/train-*
- config_name: media_shard_000004
  data_files:
  - split: train
    path: media_shard_000004/train-*
- config_name: media_shard_000005
  data_files:
  - split: train
    path: media_shard_000005/train-*
- config_name: media_shard_000006
  data_files:
  - split: train
    path: media_shard_000006/train-*
- config_name: media_shard_000007
  data_files:
  - split: train
    path: media_shard_000007/train-*
- config_name: media_shard_000008
  data_files:
  - split: train
    path: media_shard_000008/train-*
- config_name: media_shard_000009
  data_files:
  - split: train
    path: media_shard_000009/train-*
- config_name: media_shard_000010
  data_files:
  - split: train
    path: media_shard_000010/train-*
- config_name: media_shard_000011
  data_files:
  - split: train
    path: media_shard_000011/train-*
- config_name: media_shard_000012
  data_files:
  - split: train
    path: media_shard_000012/train-*
- config_name: media_shard_000013
  data_files:
  - split: train
    path: media_shard_000013/train-*
- config_name: media_shard_000014
  data_files:
  - split: train
    path: media_shard_000014/train-*
- config_name: media_shard_000015
  data_files:
  - split: train
    path: media_shard_000015/train-*
- config_name: media_shard_000016
  data_files:
  - split: train
    path: media_shard_000016/train-*
- config_name: media_shard_000017
  data_files:
  - split: train
    path: media_shard_000017/train-*
- config_name: queries
  data_files:
  - split: train
    path: queries/train-*
---

# OmniRet training dataset

`OmniRet-train` is the training-data release for
[OmniRet](https://arxiv.org/abs/2603.02098), a unified retrieval model for
text, image, video, and audio. This card documents the released snapshot for
researchers training or analyzing OmniRet.

## Dataset summary

The release contains 6,405,109 query rows and 7,119,841 candidate rows from 30
datasets. It covers 15 retrieval directions across text (T), image (I), video
(V), and audio (A). The OmniRet paper reports this corpus as approximately 6.4
million query-candidate pairs.

## Provenance

OmniRet starts from the
[M-BEIR](https://huggingface.co/datasets/TIGER-Lab/M-BEIR) universal retrieval
benchmark and extends it with text, image-text, video-text, audio-text, and
audio-visual training data. The source groups follow the training-data section
and Table 10 of the [OmniRet paper](https://arxiv.org/abs/2603.02098).

| Source group | Datasets |
| --- | --- |
| M-BEIR foundation | NIGHTS, WebQA, VisualNews, Fashion200K, MSCOCO, EDIS, OVEN, InfoSeek, FashionIQ, CIRR |
| Added text retrieval | MS MARCO, HotpotQA, Natural Questions, PAQ, StackExchange, NLI, SQuAD |
| Added image-text and composed image retrieval | LLaVA-558K, CC-CoIR, MTCIR |
| Added video-text and composed video retrieval | TGIF, Charades, WebVid2M, PE-Video, WebCoVR |
| Added audio-text retrieval | AudioCaps, Clotho v2.1, WavText5K, WavCaps |
| Added audio-visual retrieval | VGGSound |

## Retrieval tasks

| Family | Directions |
| --- | --- |
| Unimodal | I → I; T → T |
| Cross-modal binding | I → T; T → I; V → T; T → V; A → T; T → A |
| Composed retrieval | T → (I,T); (I,T) → T; (I,T) → I; (I,T) → (I,T); (V,T) → V |
| Audio-visual binding | A → V; V → A |

## Repository layout

| Path | Contents |
| --- | --- |
| `queries/` | Training queries, task IDs, positive candidate IDs, text, and media references |
| `candidates/` | Candidate records referenced by the queries |
| `instructions/` | Retrieval instructions keyed by task and dataset |
| `media_shard_*/` | Sharded image, video, and audio payloads |
| `metadata/candidate_index.sqlite` | Read-only lookup index for 7,119,841 candidates |
| `metadata/media_index.sqlite` | Read-only lookup index for 3,538,323 media IDs across 1,976 Parquet files |
| `metadata/train_pairs.jsonl` | Precomputed manifest containing 6,405,109 training pairs |

Download the repository to the canonical training path:

```bash
hf download chuonghm/OmniRet-train \
  --repo-type dataset \
  --local-dir /data1/omniret/OmniRet-train-data
```

The published media index stores Parquet paths below
`/data1/omniret/OmniRet-train-data`. Use that location, or provide a compatible
media index when using another root.

## Relationship to the ACM benchmark

The OmniRet paper also introduces the separate
[Audio-Centric Multimodal benchmark (ACM)](https://huggingface.co/datasets/chuonghm/ACM),
curated from VGGSound. ACM evaluates composed audio retrieval (A,T → A) and
bidirectional audio-image and audio-video retrieval (A → I, I → A, A → V,
V → A). Use `chuonghm/ACM` for evaluation; this training repository does not
replace the benchmark release.

## Data terms

This repository is a transformed compilation of upstream datasets. Underlying
examples remain subject to their original licenses and terms. Review the
[M-BEIR dataset card](https://huggingface.co/datasets/TIGER-Lab/M-BEIR) and the
source references in Table 10 of the
[OmniRet paper](https://arxiv.org/abs/2603.02098) before redistribution or
commercial use.

## Citation

Please cite OmniRet for this extended training release and ACM benchmark, and
cite UniIR for the M-BEIR foundation.

```bibtex
@article{huynh2026omniret,
  title   = {Efficient and High-Fidelity Omni Modality Retrieval},
  author  = {Huynh, Chuong and Luong, Manh and Shrivastava, Abhinav},
  journal = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year    = {2026}
}

@article{wei2023uniir,
  title   = {UniIR: Training and Benchmarking Universal Multimodal Information Retrievers},
  author  = {Wei, Cong and Chen, Yang and Chen, Haonan and Hu, Hexiang and Zhang, Ge and Fu, Jie and Ritter, Alan and Chen, Wenhu},
  journal = {arXiv preprint arXiv:2311.17136},
  year    = {2023}
}
```