Vietnamese Visual Question Answering Dataset With 37,000+ QA Pairs

OpenViVQA provides over 11,000 images paired with more than 37,000 open-ended question-answer pairs in Vietnamese. The dataset was created by uitnlp for the VLSP 2023 - ViVRC shared task challenge and was last updated in December 2023.

Use Cases

• Train models to answer open-ended Vietnamese questions about image content using the provided QA pairs.
• Benchmark model performance on Vietnamese visual reasoning by evaluating on the private test set.
• Fine-tune vision-language models for the Vietnamese domain using the image-text pairs.
• Study cross-lingual transfer by comparing performance on this dataset to English VQA benchmarks.

Strengths

• Contains over 37,000 question-answer pairs for model training and evaluation.
• Includes more than 11,000 unique images as a visual foundation.

Limitations

• Dataset size is moderate; 37,000 QA pairs may be insufficient for training large models from scratch without augmentation or transfer learning.
• Focus is exclusively on Vietnamese language and culture, limiting generalizability to other languages without adaptation.

Provenance

Source

uitnlp, associated with the VLSP 2023 - ViVRC shared task.

Collection Method

Created for a shared task challenge; specific collection method is null.

Time Range

null

Freshness

Last updated in December 2023.

Geography

null