Scholarlink

The present study focuses on the development of MLOPs pipelines for the accurate prediction of fat percentage in beef images. Using two U-Net models from our previous work, one for background removal and one for fat segmentation, the system achieves high accuracy in image analysis. At the same time, an unsupervised autoencoder model in combination with a SVM decision score is applied to check the incoming data, ensuring that the images processed by the system are suitable for analysis. To monitor and automate the process, tools such as MLflow are used, which monitors model performance and triggers retraining when needed, and GitHub Actions, which facilitates automatic integration and delivery of new model versions to production. This system provides an integrated and automated framework that is capable of dynamically adapting to changes in data and maintaining high levels of performance in production environments, the study suggests a path to accurately and efficiently implement MLOps pipelines, using the latest machine learning and automation technologies.

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

Toward the Reliable Deployment of Computer Vision Pipelines by Leveraging MLOps and Explainable AI

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Evangelos Nerantzis,
  • Georgios Symeonidis,
  • Aikaterini Vouta Papageorgiou,
  • George A. Papakostas

摘要

The present study focuses on the development of MLOPs pipelines for the accurate prediction of fat percentage in beef images. Using two U-Net models from our previous work, one for background removal and one for fat segmentation, the system achieves high accuracy in image analysis. At the same time, an unsupervised autoencoder model in combination with a SVM decision score is applied to check the incoming data, ensuring that the images processed by the system are suitable for analysis. To monitor and automate the process, tools such as MLflow are used, which monitors model performance and triggers retraining when needed, and GitHub Actions, which facilitates automatic integration and delivery of new model versions to production. This system provides an integrated and automated framework that is capable of dynamically adapting to changes in data and maintaining high levels of performance in production environments, the study suggests a path to accurately and efficiently implement MLOps pipelines, using the latest machine learning and automation technologies.