MIRACL-VISION

MIRACL-VISION is a multilingual visual retrieval dataset for 18 different languages. It is an extension of MIRACL, a popular text-only multilingual retrieval dataset. The dataset contains user questions, images of Wikipedia articles and annotations, which article can answer a user question. There are 7898 questions and 338734 images. More details can be found in the paper MIRACL-VISION: A Large, multilingual, visual document retrieval benchmark.

This dataset is ready for commercial usage for evaluation of the multilingual, multimodal retriever pipelines.

Correspondence to

Benedikt Schifferer (bschifferer@nvidia.com)

Dataset Creation Date:

31st January 2025

License/Terms of Use:

This dataset is licensed under Creative Commons Attribution-ShareAlike 4.0 International. Additional Information: Apache License 2.0.

Intended Usage:

Users can evaluate multilingual, multimodal retriever pipelines.

Dataset Characterization

Dataset Collection Method: Automated Labelling Method: Human

Dataset Format

The images are stored Pillow (PIL) Images in HuggingFace Dataset format The questions, corpus, questions-corpus pairs are stored in parquet/BEIR format

Reference(s):

@misc{osmulsk2025miraclvisionlargemultilingualvisual,
      title={MIRACL-VISION: A Large, multilingual, visual document retrieval benchmark}, 
      author={Radek Osmulsk and Gabriel de Souza P. Moreira and Ronay Ak and Mengyao Xu and Benedikt Schifferer and Even Oldridge},
      year={2025},
      eprint={2505.11651},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2505.11651}, 
}

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Please report security vulnerabilities or NVIDIA AI Concerns here.

Example

The requirements are

pip install qwen_vl_utils beir==2.0.0

The dataset contains an eval_example using MrLight/dse-qwen2-2b-mrl-v1​​

python embedding_eval.py --dataset nvidia/miracl-vision --language en

Loading the Dataset

from datasets import load_dataset

def hf_beir_queries(queries):
    queries_beir = {}
    for query in queries:
        queries_beir[query['_id']] = query['text']
    return(queries_beir)

def hf_beir_corpus(corpus):
    corpus_beir = {}
    for doc in corpus:
        corpus_beir[doc['_id']] = doc
    return(corpus_beir)

def hf_beir_qrels(qrels):
    qrels_beir = {}
    for el in qrels:
        if str(el['query-id']) in qrels_beir:
            qrels_beir[str(el['query-id'])][str(el['corpus-id'])] = el['score']
        else:
            qrels_beir[str(el['query-id'])] = {str(el['corpus-id']): el['score']}
    return(qrels_beir)

def load_data(
    path,
    lang
):
    queries = load_dataset(path, 'queries-' + str(lang), split='default')
    queries = hf_beir_queries(queries)
    corpus = load_dataset(path, 'corpus-' + str(lang), split='default')
    corpus = hf_beir_corpus(corpus)
    qrels = load_dataset(path, 'qrels-' + str(lang), split='default')
    qrels = hf_beir_qrels(qrels)
    images = load_dataset(path, 'images-' + str(lang), split='default')
    return(queries, corpus, qrels, images)

queries, corpus, qrels, images = load_data('nvidia/miracl-vision', 'en')

Dataset Statistics

Number of Images: 338734

Number of questions: 7898

Total Data Storage: 95GB

		MIRACL (original)	MIRACL (original)	MIRACL-VISION	MIRACL-VISION
Language	Language Code	# of queries	# of document chunks	# of queries	# of documents
Arabic	ar	2896	2061414	2127	75444
Bengali	bn	411	297265	229	8495
Chinese	zh	393	4934368	189	8672
English	en	799	32893221	447	42971
Farsi	fa	632	2207172	342	15846
Finnish	fi	1271	1883509	791	33679
French	fr	343	14636953	142	6990
German	de	305	15866222	129	6302
Hindi	hi	350	506264	184	8004
Indonesian	id	960	1446315	603	23842
Japanese	ja	860	6953614	387	17909
Korean	ko	213	1486752	130	5700
Russian	ru	1252	9543918	564	25201
Spanish	es	648	10373953	369	17749
Swahili	sw	482	131924	239	7166
Telugu	te	828	518079	480	15429
Thai	th	733	542166	451	16313
Yoruba	yo	119	49043	95	3022
Avereage		750	5907342	439	18819

Results

	MIRACL-VISION (Text)	MIRACL-VISION (Text)	MIRACL-VISION (Text)	MIRACL-VISION (Text)	MIRACL-VISION (Image)	MIRACL-VISION (Image)	MIRACL-VISION (Image)	MIRACL-VISION (Image)
	multilingual-e5-large	snowflake-arctic-embed-l-v2.0	gte-multilingual-base	bge-m3	dse-qwen2-2b-mrl-v1	gme-Qwen2-VL-2B-Instruct	vdr-2b-multi-v1	colqwen2-v1.0
LLM Parameters (in M)	560	567	305	567	1543	1543	1543	1543
Language
Arabic	0.8557	0.8754	0.8503	0.8883	0.3893	0.4888	0.4379	0.4129
Bengali	0.8421	0.8325	0.8211	0.8585	0.2352	0.3755	0.2473	0.2888
Chinese	0.6900	0.7179	0.7167	0.7458	0.5962	0.6314	0.5963	0.4926
English	0.7029	0.7437	0.7345	0.7348	0.6605	0.6784	0.6784	0.6417
Farsi	0.6793	0.7001	0.6984	0.7297	0.2250	0.3085	0.2398	0.2616
Finnish	0.8974	0.9014	0.8957	0.9071	0.4162	0.6863	0.5283	0.6604
French	0.7208	0.8236	0.7771	0.8158	0.7160	0.6851	0.7194	0.6876
German	0.7622	0.7774	0.7498	0.7695	0.6267	0.6345	0.6205	0.5995
Hindi	0.7595	0.7255	0.6916	0.7581	0.1740	0.3127	0.2058	0.2209
Indonesian	0.6793	0.6906	0.6757	0.7049	0.4866	0.5416	0.5254	0.5320
Japanese	0.8378	0.8484	0.8442	0.8720	0.6232	0.7305	0.6553	0.6970
Korean	0.7327	0.7545	0.7397	0.7934	0.4446	0.6202	0.4952	0.4419
Russian	0.7857	0.8242	0.8023	0.8363	0.6505	0.7202	0.6995	0.6811
Spanish	0.6596	0.7250	0.7029	0.7268	0.5927	0.6277	0.6274	0.6224
Swahili	0.8157	0.8089	0.7987	0.8337	0.4156	0.5348	0.4509	0.4931
Telugu	0.8948	0.9201	0.9076	0.9090	0.0274	0.0893	0.0318	0.0264
Thai	0.8424	0.8485	0.8509	0.8682	0.2692	0.3563	0.3177	0.2389
Yoruba	0.5655	0.5332	0.5698	0.5842	0.4178	0.4884	0.4577	0.5120
Average	0.7624	0.7806	0.7682	0.7964	0.4426	0.5283	0.4741	0.4728
Average w/o Telugu	0.7546	0.7724	0.7600	0.7898	0.4670	0.5542	0.5002	0.4991