在近日举行的一场技术研讨会上，苹果公司与加州大学圣塔巴巴拉分校（UCSB）的研究人员共同发布了一款名为MGIE的图片编辑框架。MGIE，即MLLM-Guided Image Editing，是一款将多模态大模型MLLM应用于图像编辑领域的创新技术。
据了解，MGIE的主要创新之处在于解决了传统图像编辑指令引导不足的问题。通过学习，MLLM可以生成简明的表达指令，并为图像编辑提供明确的视觉相关引导。在端到端训练过程中，扩散模型会同步更新，从而实现对预期目标的潜在想象力执行图像编辑。
在人类指令的引导下，MGIE可以进行Photoshop风格的修改、全局照片优化和局部对象修改。这意味着，用户只需给出简单的指令，MGIE就能自动完成复杂的图像编辑任务，极大地提高了图像编辑的效率和准确性。
苹果公司与UCSB的研究人员选择将MGIE开源，旨在推动图像编辑技术的发展，并让更多人受益于这一创新技术。据悉，MGIE的源代码已公布在 GitHub 上，感兴趣的开发者可以随时查看和使用。
英文翻译：
Title: Apple and UCSB Release New Image Editing Framework MGIE
Keywords: Apple, UCSB, MGIE, image editing, open source
News content:
At a recent technical symposium, researchers from Apple and the University of California, Santa Barbara (UCSB) jointly released a new image editing framework called MGIE. MGIE, or MLLM-Guided Image Editing, is an innovative technology that applies multimodal large model MLLM to the field of image editing.
It is understood that the main innovation of MGIE lies in solving the problem of insufficient instruction guidance in traditional image editing. By learning, MLLM can generate concise expression instructions and provide clear visual guidance for image editing. In the end-to-end training process, the diffusion model will be updated synchronously, thereby realizing the potential imagination of the expected target for image editing.
Under the guidance of human instructions, MGIE can perform Photoshop-style modifications, global photo optimization, and local object modifications. This means that users only need to give simple instructions, and MGIE can automatically complete complex image editing tasks, greatly improving the efficiency and accuracy of image editing.
Apple and UCSB researchers choose to open source MGIE, with the aim of promoting the development of image editing technology and benefiting more people with this innovative technology. It is reported that the source code of MGIE has been published on GitHub, and interested developers can view and use it at any time.
【来源】https://www.jiqizhixin.com/articles/2024-02-05-10
Views: 1