苹果公司和加州大学圣塔巴巴拉分校（UCSB）的研究人员近日共同发布了一款名为 MGIE（MLLM-Guided Image Editing）的图片编辑框架。该框架的独特之处在于，它采用了多模态大模型 MLLM 來解决指令引导不足的问题。通过学习，MLLM 能够获得简明的表达指令，并为图像编辑提供明确的视觉引导。
MGIE 框架利用端到端训练，使扩散模型能够同步更新，并利用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 具备了进行 Photoshop 风格的修改、全局照片优化和局部对象修改的能力。
这一创新的图片编辑框架有望在摄影、设计等领域带来广泛的应用。此次苹果公司开源 MGIE，也将有助于推动相关技术的发展和普及。
英文翻译：
News Title: Apple Releases Open-Source Image Editing Framework
Keywords: Apple, Open-source, Image Editing
News Content:
Apple and researchers from the University of California, Santa Barbara (UCSB) have jointly developed a groundbreaking image editing framework called MGIE (MLLM-Guided Image Editing). The framework utilizes multimodal large model MLLM to address the issue of insufficient instruction guidance. By learning, MLLM can obtain concise expression instructions and provide clear visual guidance for image editing.
Through end-to-end training, the diffusion model in the MGIE framework synchronizes updates, utilizing the potential imagination of the expected target to perform image editing. Under the guidance of human instructions, MGIE is capable of performing Photoshop-style modifications, global photo optimization, and local object modifications. This innovative image editing framework is expected to have widespread applications in photography, design, and other fields. By open-sourcing MGIE, Apple aims to promote the development and popularization of related technologies.
【来源】https://www.jiqizhixin.com/articles/2024-02-05-10
Views: 1