---
language:
- en
license: mit
task_categories:
- image-text-to-text
size_categories:
- 1M<n<10M
pretty_name: mmE5 labeled data
tags:
- embedding
- multimodal
configs:
- config_name: TAT-DQA
  data_files:
  - split: train
    path: TAT-DQA/TAT-DQA.parquet
- config_name: ArxivQA
  data_files:
  - split: train
    path: ArxivQA/ArxivQA.parquet
- config_name: InfoSeek_it2t
  data_files:
  - split: train
    path: InfoSeek_it2t/InfoSeek_it2t.parquet
- config_name: InfoSeek_it2it
  data_files:
  - split: train
    path: InfoSeek_it2it/InfoSeek_it2it.parquet
- config_name: ImageNet_1K
  data_files:
  - split: train
    path: ImageNet_1K/ImageNet_1K.parquet
- config_name: N24News
  data_files:
  - split: train
    path: N24News/N24News.parquet
- config_name: HatefulMemes
  data_files:
  - split: train
    path: HatefulMemes/HatefulMemes.parquet
- config_name: SUN397
  data_files:
  - split: train
    path: SUN397/SUN397.parquet
- config_name: VOC2007
  data_files:
  - split: train
    path: VOC2007/VOC2007.parquet
- config_name: InfographicsVQA
  data_files:
  - split: train
    path: InfographicsVQA/InfographicsVQA.parquet
- config_name: ChartQA
  data_files:
  - split: train
    path: ChartQA/ChartQA.parquet
- config_name: A-OKVQA
  data_files:
  - split: train
    path: A-OKVQA/A-OKVQA.parquet
- config_name: DocVQA
  data_files:
  - split: train
    path: DocVQA/DocVQA.parquet
- config_name: OK-VQA
  data_files:
  - split: train
    path: OK-VQA/OK-VQA.parquet
- config_name: Visual7W
  data_files:
  - split: train
    path: Visual7W/Visual7W.parquet
- config_name: VisDial
  data_files:
  - split: train
    path: VisDial/VisDial.parquet
- config_name: CIRR
  data_files:
  - split: train
    path: CIRR/CIRR.parquet
- config_name: NIGHTS
  data_files:
  - split: train
    path: NIGHTS/NIGHTS.parquet
- config_name: WebQA
  data_files:
  - split: train
    path: WebQA/WebQA.parquet
- config_name: VisualNews_i2t
  data_files:
  - split: train
    path: VisualNews_i2t/VisualNews_i2t.parquet
- config_name: VisualNews_t2i
  data_files:
  - split: train
    path: VisualNews_t2i/VisualNews_t2i.parquet
- config_name: MSCOCO_i2t
  data_files:
  - split: train
    path: MSCOCO_i2t/MSCOCO_i2t.parquet
- config_name: MSCOCO_t2i
  data_files:
  - split: train
    path: MSCOCO_t2i/MSCOCO_t2i.parquet
- config_name: MSCOCO
  data_files:
  - split: train
    path: MSCOCO/MSCOCO.parquet
---

# mmE5 Labeled Data

This dataset contains datasets used for the supervised finetuning of mmE5 ([mmE5: Improving Multimodal Multilingual Embeddings via High-quality Synthetic Data](https://huggingface.co/papers/2502.08468)):
- **MMEB** (with hard negative)
- **InfoSeek** (from M-BEIR)
- **TAT-DQA**
- **ArxivQA**

[Github](https://github.com/haon-chen/mmE5)

## Image Preparation

First, you should prepare the images used for training:

### Image Downloads

- **Download Links**: Download image resources for each dataset via the following links:
  - [**MMEB**](https://huggingface.co/datasets/TIGER-Lab/MMEB-train)
  - [**InfoSeek**](https://huggingface.co/datasets/TIGER-Lab/M-BEIR)
  - [**ArxivQA**](https://huggingface.co/datasets/MMInstruction/ArxivQA)
  - [**TAT-DQA**](https://huggingface.co/datasets/vidore/tatdqa_train)

For TAT-DQA, you need to first save images into the overall image folder to align usage:

```
dataset = load_dataset(
    "vidore/tatdqa_train",
    split="train"
)
image_out_dir = "images/TAT-DQA"
os.makedirs(image_out_dir, exist_ok=True)
for i, sample in enumerate(dataset):
  save_path = os.path.join(image_out_dir, f"tatdqa_{i}.png")
  if os.path.exists(save_path):
    continue
  image = sample["image"]
  image.save(save_path, format="PNG")
```

### Image Organization

```
  images/
  ├── mbeir_images/
  │     └── oven_images/
  │           └── ... .jpg (InfoSeek)
  ├── ArxivQA/
  │     └── images/
  │           └── ... .jpg (ArxivQA)
  └── TAT-DQA/
  │     └── ... .png (TAT-DQA)
  └── A-OKVQA/
        └── Train/
  │           └── ... .jpg (A-OKVQA)
  │
  ... (MMEB Training images)
```

You can refer to the image paths in each subset to view the image organization.

You can also customize your image paths by altering the image_path fields.
## Citation
If you use this dataset in your research, please cite the associated paper.
```
@article{chen2025mmE5,
  title={mmE5: Improving Multimodal Multilingual Embeddings via High-quality Synthetic Data},
  author={Chen, Haonan and Wang, Liang and Yang, Nan and Zhu, Yutao and Zhao, Ziliang and Wei, Furu and Dou, Zhicheng},
  journal={arXiv preprint arXiv:2502.08468},
  year={2025}
}
```