A Comparative Study on the Performance of Generative Models for Augmenting Red Pepper Crop Data: Focusing on DDIM, DDPM, VQ-VAE, and WGAN-GP

Red peppers are susceptible to diseases and pests, necessitating effective management strategies. This paper compares the performance of four generative models—DDIM, DDPM, VQ-VAE, and WGAN-GP—for generating red pepper crop images. Evaluation using FID, LPIPS, and PSNR metrics revealed that DDIM exhibited superior performance in image quality and diversity, while VQ-VAE excelled in reconstruction accuracy. DDPM showed balanced performance across metrics, whereas WGAN-GP demonstrated relatively lower performance. Qualitative assessment further confirmed that DDIM and VQ-VAE most accurately reproduced the detailed features of red pepper crops. These findings provide crucial insights for disease diagnosis and data augmentation in red pepper cultivation, highlighting the potential applications of generative models in agriculture. The study contributes to informed decision-making in crop management and opens avenues for improving agricultural productivity and quality through advanced image generation techniques.