Indo-CXR-VQA: Indonesian Visual Question Answering for Chest X-Rays

Indonesian-language Visual Question Answering dataset derived from VinDr-CXR radiologist annotations. It contains 15,991 question–answer–reason triples grounded in annotated findings. The dataset was created by Softcase and was last updated on 2026-07-09.

Use Cases

• Train Visual Question Answering models based on Indonesian-language questions about chest X-rays.
• Develop models for generating radiological reasoning based on question–answer–reason triples.
• Benchmark semantic VQA performance on medical images using the 'noloc' configuration.
• Build separate grounding models for medical image localization using data with bounding box annotations.

Strengths

• Contains 15,991 examples of question–answer–reason triples.
• Data is grounded in single radiologist-annotated findings, suggesting expert-level labels.
• Offers a specific 'noloc' configuration for semantic VQA, with localization handled separately.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known only for the default configuration; total dataset scale is unclear.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Derived from VinDr-CXR radiologist annotations and VinDr-CXR-VQA.

Collection Method

Likely created by processing expert annotations to generate question–answer–reason triples.

Freshness

Last updated 2026-07-09 04:09:14.

Geography

The dataset uses Indonesian language, suggesting a focus on Indonesia.