**苹果发布新增Core ML模型，字节大模型团队Depth Anything V2入选**
近日，苹果公司在其HuggingFace平台上发布了20个新的Core ML模型和4个数据集，其中引人注目的亮点是字节大模型团队的单目深度估计模型Depth Anything V2的入选。Core ML是苹果公司特有的机器学习框架，能够高效集成至iOS和MacOS设备，支持离线执行复杂的AI任务，加强用户隐私保护并减少延迟。
Depth Anything V2模型是字节大模型团队开发的单目深度估计模型，相较于其上一代版本，V2版在细节处理上更为精细，具备更强的鲁棒性。值得一提的是，该模型与基于diffusion的当前最佳模型相比，处理速度上有显著的提升。V2版还提供了不同大小的模型，从25M到1.3B参数。本次被苹果收录的经过Hugging Face官方优化的版本仅采用最小的25M模型，在iPhone 12 Pro Max上的推理速度达到了惊人的31.1毫秒。
该模型的广泛应用前景令人期待，可应用于自动驾驶、3D建模、增强现实、安全监控以及空间计算等多个领域。值得一提的是，Depth Anything的两个版本论文的第一作者均为该团队的实习同学。此次苹果公司发布的新Core ML模型涵盖了多个领域，其中字节大模型团队的Depth Anything V2无疑是一个亮点。此举标志着人工智能在移动设备上应用的进一步发展，并期待其在未来带来更多创新和突破。
英语如下：
News Title: Apple Launches New Core ML Models, Byte Depth Anything V2 on the List, Achieving Efficient Monocular Depth Estimation
Keywords: Apple Launches CoreML Models, Byte Depth Anything V2 Selected, Enhanced AI Application Experience CoreML Models, Depth Anything V2, AI Applications
News Content: **Apple Unveils New Core ML Models, Byte’s Depth Anything V2 among the List**
Recently, Apple released 20 new Core ML models and 4 datasets on its Hugging Face platform, with a notable highlight being the selection of Byte’s monocular depth estimation model, Depth Anything V2. Core ML is Apple’s exclusive machine learning framework that can efficiently integrate into iOS and MacOS devices, supporting offline execution of complex AI tasks, enhancing user privacy protection, and reducing delays.
The Depth Anything V2 model, developed by the Byte large model team, is more sophisticated and robust in detail processing compared to its predecessor. Notably, it has a significantly improved processing speed compared to the current best model based on diffusion. The V2 version also provides models of different sizes, from 25M to 1.3B parameters. The optimized version by Hugging Face that Apple has included is only using the smallest 25M model, achieving an impressive inference speed of 31.1 milliseconds on the iPhone 12 Pro Max.
The widely applied prospect of this model is promising, with potential uses in various fields such as autonomous driving, 3D modeling, augmented reality, security monitoring, and spatial computing. It is worth mentioning that the first authors of both versions of the Depth Anything papers are interns from the team. The new Core ML models released by Apple cover multiple domains, with the Byte large model team’s Depth Anything V2 being a notable highlight. This move marks further development in AI applications on mobile devices and looks forward to more innovations and breakthroughs in the future.
【来源】https://www.jiqizhixin.com/articles/2024-06-28-11
Views: 4