Transferring style from one image to another is an important utility in today’s world. There has been a lot of new research coming up that is progressing toward achieving perfection in their proposed solutions to this novel problem. However, the current research has still not been able to achieve good accuracy across images of any arbitrary domain. As well as their solutions require decent hardware configuration to run on which makes it cost-ineffective to have their solutions deployed in real-world settings. In this paper, we propose an architecture that can be used to deploy such models in real-world settings. We also demonstrate the results after using the proposed architecture and deploying one such model (AdaIN) that is one of the fastest for the job of transferring style across images. Prior to deployment of AdaIN, we configure it with settings that make AdaIN able to run on commodity hardware configuration with a slight drop in accuracy of the model. We aim to provide readymade architecture for researchers to deploy their models. This should enable others to try this research and hence encourage them to explore and provide their solutions to this novel problem of transferring style across images with any arbitrary domains.

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

Architecture for Cost-Effective Deployment of Models to Transfer Style Across Images

  • Niyaj Nadaf,
  • Pranati Waghodekar,
  • Aditya Magdum,
  • Priyanka Gupta,
  • Vishal Kisan Borate,
  • Yogesh Kisan Mali

摘要

Transferring style from one image to another is an important utility in today’s world. There has been a lot of new research coming up that is progressing toward achieving perfection in their proposed solutions to this novel problem. However, the current research has still not been able to achieve good accuracy across images of any arbitrary domain. As well as their solutions require decent hardware configuration to run on which makes it cost-ineffective to have their solutions deployed in real-world settings. In this paper, we propose an architecture that can be used to deploy such models in real-world settings. We also demonstrate the results after using the proposed architecture and deploying one such model (AdaIN) that is one of the fastest for the job of transferring style across images. Prior to deployment of AdaIN, we configure it with settings that make AdaIN able to run on commodity hardware configuration with a slight drop in accuracy of the model. We aim to provide readymade architecture for researchers to deploy their models. This should enable others to try this research and hence encourage them to explore and provide their solutions to this novel problem of transferring style across images with any arbitrary domains.