SalArt-VQA: A Benchmark for Salient Artifact Understanding in AI-Generated Images

950 test rows comprise the SalArt-VQA benchmark for visual question answering focused on salient artifacts in AI-generated images. The dataset includes 475 artifact images, 356 clean real-image references, and 119 paired generated artifact-free counterparts. It was created by salartvqa and last updated on Hugging Face in May 2026.

Use Cases

• Benchmarking VQA model performance on detecting artifacts in AI-generated images based on the described artifact images.
• Training models to distinguish between AI-generated images with artifacts and their clean counterparts based on the 119 paired samples.
• Studying the types and salience of artifacts produced by image generation models based on the closed-set question-answer format.
• Developing artifact mitigation techniques by comparing generated images against the 356 clean real-image references.

Strengths

• Contains 475 specifically curated artifact images for focused evaluation.
• Includes 119 paired artifact-free counterparts, enabling controlled comparative analysis.
• Provides 356 clean real-image references as a baseline for artifact-free content.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Dataset size is limited to a test split of 950 rows, which may restrict training applications.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Freshness

Last updated 2026-05-04 21:26:20; freshness should be verified.