Multi Domain VQA 20K: A Visual Question Answering Benchmark for VLMs

20,000 samples combine questions and images from three established VQA datasets: AOKVQA, Path-VQA, and TDIUC. This medium-sized benchmark is designed to test the multi-domain knowledge of vision-language models. It was created by dutta18 for educational and research purposes, with copyright retained by the original dataset owners.

Use Cases

• Benchmarking model performance on multi-domain visual question answering based on the combined AOKVQA, Path-VQA, and TDIUC splits.
• Testing model knowledge on quantitative and physical reasoning questions based on the TDIUC subset.
• Rapid prototyping of VLM inference pipelines using a pre-consolidated dataset.
• Educational demonstrations of VLM capabilities across different question and image domains.

Strengths

• Consolidates 20,000 samples from three established VQA benchmarks for multi-domain testing.
• Specifically includes quantitative and physical reasoning questions from the TDIUC validation split.
• Last updated on 2025-07-19, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect the geographic, temporal, or source biases inherent to the original constituent datasets.

Provenance

Source

Combines samples from AOKVQA, Path-VQA, and TDIUC datasets.

Collection Method

Created by dutta18 by merging train and validation splits from the source datasets.

Freshness

Last updated 2025-07-19 12:13:51.