Viet-ViTextVQA-gemini-VQA: Vietnamese Image VQA Annotations from Gemini

9,594 Vietnamese images from the ViTextVQA train split were analyzed using the Gemini 1.5 Flash model to generate over 50,000 detailed descriptions, questions, and answers. The dataset was created by 5CD-AI and was last updated on August 25, 2024. It is hosted on the Hugging Face platform.

Use Cases

• Fine-tuning VQA models for Vietnamese language based on the described image-question-answer triplets.
• Benchmarking model performance on Vietnamese visual understanding tasks using the annotated images.
• Training or evaluating multimodal large language models (MLLMs) on a non-English visual dataset.
• Studying cross-lingual transfer learning for VQA using a dataset focused on a specific language.

Strengths

• Contains 9,594 Vietnamese images, providing a focused linguistic and visual context.
• Includes over 50,000 detailed annotations generated by a leading model (Gemini 1.5 Flash).
• Based on a known dataset split (ViTextVQA train split), suggesting a structured foundation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the annotations is unknown, which may limit suitability assessment.
• Data may reflect geographic or linguistic bias inherent to the source images.

Provenance

Source

Images sourced from the train split of the ViTextVQA dataset.

Collection Method

Annotations generated using the Gemini 1.5 Flash model via advanced VQA techniques.

Freshness

Last updated 2024-08-25 15:08:02; freshness should be verified.

Geography

Vietnam (based on image source description)