苹果公司与加州大学圣塔巴巴拉分校（UCSB）的研究人员合作，推出了一款名为MGIE的图片编辑框架。MGIE（MLLM-Guided Image Editing）的核心技术创新之处在于，它利用多模态大模型MLLM解决了传统图片编辑工具在指令引导上的不足。MLLM能够学习并生成简洁的指令表达，同时提供明确的视觉引导。
这款框架通过端到端的训练，使扩散模型能够同步更新，并利用预期目标的潜在想象力来执行图像编辑任务。在人类指令的引导下，MGIE能够进行类似于Photoshop的图像修改，全局照片优化以及局部对象的精细调整。这一技术有望为图片编辑领域带来革命性的变革。
MGIE的开源特性意味着更多的开发者可以访问并参与到这个项目的开发和完善中来，有望推动整个图像编辑领域的技术进步。这一举措也展示了苹果公司在技术共享和开源社区合作方面的开放态度。
英文翻译：
Apple and UCSB Researchers Launch Open-Source Image Editing Framework MGIE
Keywords: Apple, UCSB, MGIE, Image Editing, Open Source
Apple and UCSB researchers have jointly developed and open-sourced a new image editing framework called MGIE. At the core of MGIE (MLLM-Guided Image Editing) is a technological innovation that leverages the multimodal large model MLLM to address the problem of insufficient instruction guidance in traditional image editing tools. MLLM is capable of learning and generating concise instruction expressions, while also providing clear visual guidance.
This framework is trained end-to-end, allowing the diffusion model to be updated synchronously, and to utilize the potential imagination of the expected target to perform image editing tasks. Under human instruction, MGIE can carry out image modifications similar to Photoshop, global photo optimization, and fine-tuning of local objects. This technology is expected to bring about a revolutionary change in the field of image editing.
The open-source nature of MGIE means that more developers can access and participate in the development and improvement of this project, which is expected to drive technological progress in the entire image editing field. This initiative also demonstrates Apple’s open attitude towards technology sharing and collaboration with the open-source community.
【来源】https://www.jiqizhixin.com/articles/2024-02-05-10
Views: 1