AI-Generated Texture Images for Tactile Roughness Evaluation Across Six Materials

5.2 GB of texture images generated by a text-to-image AI model for quantitative tactile roughness evaluation. The dataset spans six material categories—wood, metal, stone, fiber, plastic, and ceramic—with roughness levels systematically varied by numerical prompts. Created by Yui Momiyama and last updated in May 2026, it is intended to support reproducibility and research on tactile perception and AI-based material representation.

Use Cases

• Training models to predict tactile roughness from visual texture features based on AI-generated images.
• Benchmarking generative AI's ability to create perceptually meaningful material textures across categories like wood and metal.
• Studying the relationship between quantitative prompt parameters and perceived material properties.
• Developing datasets for haptic rendering and virtual material simulation based on visual inputs.

Strengths

• Systematic control of a key perceptual variable (roughness) using quantitative numerical prompts.
• Covers six distinct material categories, providing a degree of categorical diversity.
• Includes generated images, category labels, roughness levels, and prompt metadata to support reproducibility.
• Dataset size of 5.2 GB suggests a substantial collection of high-resolution images.

Limitations

• Row count and specific image dimensions are unknown, limiting precise suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset's realism and perceptual alignment are inherently limited by the generative AI model used.

Provenance

Source

Yui Momiyama via figshare.

Collection Method

Images generated using a text-to-image generative AI model with quantitative prompt expressions.

Freshness

Last updated 2026-05-28 13:57:28; freshness should be verified.