Scholarlink

Generating new realistic images from images of multiple fruits is crucial for multiple applications such as calorie estimation, and menu design due to the fact that the availability of big food datasets that gather diverse fruits genre is a challenge. Wasserstein Generative Adversarial Networks are robust generative models that to some limit enhanced the training instability of Generative Adversarial Networks. In this paper, to create photorealistic, multi-label fruit images, Gradient Penalty is utilized for the enhancement of the Conditional Wasserstein Generative Adversarial Networks (cWGANs-GP). Using the DeepFruits Dataset which contains over 21,000 images featuring three to five kinds of fruits from 20 diverse types arranged into eight different fruit set combinations. Fréchet Inception Distance (FID) and Inception Score (IS) are the utilized evaluation metrics, together with expert assessments. The capability of cWGAN-GP is demonstrated to produce high-fidelity, semantically coherent images that reflect specified fruit categories. The latest progress in generative modeling has highlighted the significance of producing controllable images with high quality. The proposed cWGAN-GP model is consistent with these ongoing advancements. By providing a specialized solution for multi-label fruit image generation, this work advances the field of food computing and contributes to the evolving landscape of data-driven dietary assessment and personalized nutrition technology.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Calorie Estimation and Multi-label Fruit Image Synthesis Through Conditional Wasserstein GANs with Gradient Penalty

  • 章节

  • 发表日期：

    2026年

查看全文
  • Kai Xiao,
  • Rasha S. Aboul-Yazeed,
  • Ashraf Darwish,
  • Aboul Ella Hassanien

摘要

Generating new realistic images from images of multiple fruits is crucial for multiple applications such as calorie estimation, and menu design due to the fact that the availability of big food datasets that gather diverse fruits genre is a challenge. Wasserstein Generative Adversarial Networks are robust generative models that to some limit enhanced the training instability of Generative Adversarial Networks. In this paper, to create photorealistic, multi-label fruit images, Gradient Penalty is utilized for the enhancement of the Conditional Wasserstein Generative Adversarial Networks (cWGANs-GP). Using the DeepFruits Dataset which contains over 21,000 images featuring three to five kinds of fruits from 20 diverse types arranged into eight different fruit set combinations. Fréchet Inception Distance (FID) and Inception Score (IS) are the utilized evaluation metrics, together with expert assessments. The capability of cWGAN-GP is demonstrated to produce high-fidelity, semantically coherent images that reflect specified fruit categories. The latest progress in generative modeling has highlighted the significance of producing controllable images with high quality. The proposed cWGAN-GP model is consistent with these ongoing advancements. By providing a specialized solution for multi-label fruit image generation, this work advances the field of food computing and contributes to the evolving landscape of data-driven dietary assessment and personalized nutrition technology.