FGVQA: Fine-Grained Visual Question Answering Benchmark Suite

A benchmark suite introduced in the paper 'Same or Not? Enhancing Visual Perception in Vision-Language Models'. It contains 12,000 challenging (image, question, answer) tuples emphasizing fine-grained image understanding. The dataset is composed of six sub-benchmarks and is hosted by glab-caltech.

Use Cases

• Benchmarking model performance on fine-grained visual understanding based on the described six sub-benchmarks
• Training vision-language models to improve on detailed image-question-answer tasks
• Evaluating AI systems on challenging visual question answering tasks that require nuanced image analysis

Strengths

• Contains 12,000 challenging (image, question, answer) tuples
• Composed of six distinct sub-benchmarks: TWIN-eval, ILIAS, Google Landmarks v2, MET, CUB, and Inquire

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

glab-caltech

Collection Method

Introduced as a benchmark suite in an academic paper.

Freshness

Last updated 2025-12-30 10:25:04