This dataset contains datasets used for the supervised finetuning of mmE5 (mmE5: Improving Multimodal Multilingual Embeddings via High-quality Synthetic Data):
You can use the script provided in our source code to download all images used… See the full description on the dataset page:
https://huggingface.co/datasets/intfloat/mmE5-MMEB-hardneg.