近日，苹果公司与加州大学圣塔巴巴拉分校（UCSB）的研究人员共同推出了一款名为MGIE的图片编辑框架。MGIE（MLLM-Guided Image Editing）的核心创新之处在于，它利用了多模态大模型MLLM来解决指令引导不足的问题。这种模型能够通过学习生成简洁的指令表达，并为图像编辑提供明确的视觉引导。
通过端到端的训练，MGIE能够同步更新扩散模型，并利用预期目标的潜在想象力来执行图像编辑任务。在人类指令的引导下，MGIE展现出了类似于Photoshop的编辑能力，不仅能够进行全局照片优化，还能进行局部对象修改。这一框架的推出，无疑为图像编辑领域带来了新的突破。
苹果公司与UCSB的研究人员选择开源这一框架，也体现了他们对学术界和工业界共享技术的承诺。开源意味着更多的研究人员和技术爱好者可以访问并改进这一框架，从而推动整个图像编辑领域的发展。
英文翻译：
Title: Apple and UCSB Collaborate to Launch Open Source Image Editing Framework MGIE
Keywords: Apple, UCSB, MGIE, Image Editing, Open Source
News content:
Recently, Apple Inc. and researchers from the University of California, Santa Barbara (UCSB) have jointly developed a new image editing framework called MGIE. The core innovation of MGIE (MLLM-Guided Image Editing) lies in its use of a multimodal large model MLLM to address the issue of insufficient instruction guidance. This model is capable of generating concise instruction expressions through learning and providing clear visual guidance for image editing.
With end-to-end training, MGIE can synchronously update the diffusion model and leverage the potential imagination of the expected target to execute image editing tasks. Under the guidance of human instructions, MGIE has demonstrated editing capabilities similar to Adobe Photoshop, including global photo optimization and local object modification. The introduction of this framework represents a significant breakthrough in the field of image editing.
Apple and UCSB researchers choosing to open-source this framework reflects their commitment to sharing technology with the academic and industrial communities. Open-sourcing allows more researchers and technology enthusiasts to access and improve this framework, thereby driving the development of the entire image editing field.
【来源】https://www.jiqizhixin.com/articles/2024-02-05-10
Views: 1