Pixverve 95K: Multimodal Image-Caption Dataset for Vision-Language Models

A collection of approximately 95,000 high-resolution images and associated captions, hosted on ModelScope. The dataset was created by HaojunChen and includes a dedicated benchmark subset called PixVerve-Bench. It was last updated on May 7, 2026.

Use Cases

• Training image captioning models based on the provided image-caption pairs.
• Benchmarking vision-language model performance using the dedicated PixVerve-Bench subset.
• Fine-tuning generative models for text-to-image or image-to-text tasks based on the structured metadata.

Strengths

• Approximately 95,000 images provide a substantial scale for model training.
• Includes a dedicated benchmark subset (PixVerve-Bench) for standardized evaluation.
• Last updated on May 7, 2026, 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 geographic or source bias inherent to the collection platform.

Provenance

Source

HaojunChen via Hugging Face and ModelScope.

Freshness

Last updated 2026-05-07 07:24:11