HindwingLib: A library of leaf beetle hindwings generated by Stable Diffusion and ControlNet.

Area of Science:

• Entomology
• Evolutionary Biology
• Computer Vision

Background:

• Beetle hindwing datasets are crucial for studying beetle morphology and evolution.
• Current hindwing image collection faces challenges like limited samples, complex preparation, and accessibility issues.

Purpose of the Study:

• To introduce a novel method for generating diverse and high-fidelity beetle hindwing images using Stable Diffusion and ControlNet.
• To address the limitations in acquiring and accessing beetle hindwing image data for research.

Main Methods:

• Utilized Stable Diffusion and ControlNet, a machine learning technique, for image generation.
• Applied the methodology to generate images for 200 diverse leaf beetle hindwings.
• Conducted a comparative analysis using Structural Similarity Index (SSIM), Inception Score (IS), and Fréchet Inception Distance (FID) to evaluate image fidelity.

Main Results:

• The generated synthetic hindwing images demonstrated high fidelity, showing strong alignment with actual beetle hindwing data.
• Quantitative metrics (SSIM, IS, FID) confirmed the quality and realism of the synthetic images.
• A novel library of leaf beetle hindwing images was created.

Conclusions:

• The proposed Stable Diffusion and ControlNet approach effectively generates high-fidelity beetle hindwing images.
• This method overcomes challenges associated with traditional image collection, providing a valuable resource for machine learning and evolutionary studies.
• The generated library expands morphological data availability for beetles.