您现在访问的是微软AZURE全球版技术文档网站,若需要访问由世纪互联运营的MICROSOFT AZURE中国区技术文档网站,请访问 https://docs.azure.cn.

自定义视觉中的新增功能What's new in Custom Vision

了解服务中的新增功能。Learn what's new in the service. 这些内容可能包括发布说明、视频、博客文章和其他类型的信息。These items may be release notes, videos, blog posts, and other types of information. 将此页加入书签,以随时了解服务的最新信息。Bookmark this page to keep up-to-date with the service.

2020 年 10 月October 2020

自定义基础模型Custom base model

  • 有些应用程序有大量的联合训练数据,但需要分别对它们的模型进行微调;这样可以使来自不同来源、差异较小的图像获得更好的性能。Some applications have a large amount of joint training data but need to fine-tune their models separately; this results in better performance for images from different sources with minor differences. 在这种情况下,可以像往常一样使用大量的训练数据训练第一个模型。In this case, you can train the first model as usual with a large volume of training data. 然后使用请求正文中的 CustomBaseModelInfo 调用 3.4 公共预览 API 中的 TrainProject,以使用第一阶段训练的模型作为下游项目的基础模型。Then call TrainProject in the 3.4 public preview API with CustomBaseModelInfo in the request body to use the first stage trained model as the base model for downstream projects. 如果源项目和下游目标项目具有相似的图像特征,则预期可以获得更好的性能。If the source project and the downstream target project have similar images characteristics, you can expect better performance.

新的域信息New domain information

  • 自定义视觉 3.4 公共预览 API 中的从 GetDomains 返回的域信息现在包括支持的可导出平台、模型体系结构的简要描述以及压缩域的模型大小。The domain information returned from GetDomains in the Custom Vision 3.4 public preview API now includes supported exportable platforms, a brief description of model architecture, and the size of the model for compact domains.

训练分歧反馈Training divergence feedback

  • 现在,自定义视觉 3.4 公共预览 API 通过 GetIteration 调用返回 TrainingErrorDetails 。The Custom Vision Custom Vision 3.4 public preview API now returns TrainingErrorDetails from the GetIteration call. 对于失败的迭代,这显示了失败是否是由训练分歧引起的,这可以用更多更高质量的训练数据来弥补。On failed iterations, this reveals whether the failure was caused by training divergence, which can be remedied with more and higher-quality training data.

2020 年 7 月July 2020

Azure 基于角色的访问控制Azure role-based access control

  • 自定义视觉支持 Azure 基于角色的访问控制 (Azure RBAC),这是用于管理对 Azure 资源的个人访问权限的授权系统。Custom Vision supports Azure role-based access control (Azure RBAC), an authorization system for managing individual access to Azure resources. 若要了解如何管理对自定义视觉项目的访问权限,请参阅 Azure 基于角色的访问控制To learn how to manage access to your Custom Vision projects, see Azure role-based access control.

子集训练Subset training

  • 训练对象检测项目时,可以选择只对应用的标记的一个子集进行训练。When training an object detection project, you can optionally train on only a subset of your applied tags. 如果你还没有应用足够多的特定标记,但是你确实有足够多的其他标记,则可能需要这样做。You may want to do this if you haven't applied enough of certain tags yet, but you do have enough of others. 按照 C# 或 Python 的客户端库快速入门了解详细信息。Follow the Client library quickstart for C# or Python to learn more.

Azure 存储通知Azure storage notifications

  • 可以将自定义视觉项目与 Azure blob 存储队列集成,以获取项目训练/导出活动的推送通知和已发布模型的备份副本。You can integrate your Custom Vision project with an Azure blob storage queue to get push notifications of project training/export activity and backup copies of published models. 运行长时间的操作时,此功能有助于避免不断轮询服务来获取结果。This feature is useful to avoid continually polling the service for results when long operations are running. 你可以改为将存储队列通知集成到工作流中。Instead, you can integrate the storage queue notifications into your workflow. 有关详细信息,请参阅存储集成指南。See the Storage integration guide to learn more.

复制和移动项目Copy and move projects

  • 你现在可以将项目从一个自定义视觉帐户复制到其他帐户。You can now copy projects from one Custom Vision account into others. 你可能需要将项目从开发环境移至生产环境,或将项目备份到不同 Azure 区域中的帐户,以提高数据安全性。You might want to move a project from a development to production environment, or back up a project to an account in a different Azure region for increased data security. 有关详细信息,请参阅复制和移动项目指南。See the Copy and move projects guide to learn more.

2019 年 9 月September 2019

建议的标记Suggested tags

  • 自定义视觉网站上的智能标记工具为训练图像生成建议的标记。The Smart Labeler tool on the Custom Vision website generates suggested tags for your training images. 这使你可以在训练自定义视觉模型时更快地标记大量图像。This lets you label a large number of images more quickly when training a Custom Vision model. 有关如何使用此功能的说明,请参阅建议的标记For instructions on how to use this feature, see Suggested tags.

认知服务更新Cognitive Service updates

认知服务的 Azure 更新公告Azure update announcements for Cognitive Services