Cloud Computing Köln

ShapeMask: High-performance, large-scale instance segmentation with Cloud TPUs

Many of us take the ability to see the world for granted, but the everyday task of identifying objects of various shapes, colors, and sizes is a challenging feat. Yet, this type of technology is critical for a range of applications, from medical image analysis to photo editing. As part of our ongoing effort to create software that can perform useful visual tasks, we’ve developed a new image segmentation model called ShapeMask that offers a great combination of high accuracy and high scalability. In this blog we’ll look at what exactly ShapeMask is, what its advantages are, and how you can get started with it.An overview of ShapeMaskThe task that ShapeMask performs is called “instance segmentation,” which involves identifying and tracing the boundaries of specific instances of various objects in a visual scene. For example, in a cityscape image that contains several cars, ShapeMask can be used to highlight each car with a different color. Each of these highlighted areas is called a “mask.”ShapeMask builds on a well-known object detection model called RetinaNet (this Cloud TPU tutorial has more information) which can detect the location and size of various objects in an image but does not produce object masks. ShapeMask initially locates objects using RetinaNet, but then gradually refines the shapes of these detected objects by grouping pixels that have a similar appearance. This new approach allows ShapeMask to create accurate masks. We have made a well-optimized implementation of the ShapeMask model available open-source here.A pictorial overview of how ShapeMask starts with detection boxes and gradually refines object shapes (source of the image)A highly scalable solutionMany businesses depend on automated image segmentation to enable a broad set of applications. These businesses often work with large, frequently changing datasets, and their researchers and engineers need to experiment with a variety of ML model architectures. To iterate quickly on large, realistic datasets, they need to be able to scale up the training of their image segmentation models. One advantage ShapeMask has over other image segmentation models is that it can train efficiently at large batch sizes, which makes it possible to distribute ShapeMask training across a large number of ML accelerators.Cloud TPUs are designed to enable exactly the type of scaling that ShapeMask uses. For example, a machine perception engineer can experiment with ShapeMask on a small dataset on a single Cloud TPU v3 device—which has eight cores—and then use the same code to quickly train the same ShapeMask model on a much larger dataset using a “slice” of a Cloud TPU Pod with 32, 128, or 256 cores. Without any code changes, ShapeMask can scale to a batch size of 2048, while still achieving an accuracy of 34.7 mask mAP (more on mAP here). With a 256-core Cloud TPU v3 slice, the ShapeMask model can be trained on the standard COCO image segmentation dataset in just under 40 minutes—that’s a big improvement from waiting hours to train ShapeMask (or a comparable Mask R-CNN model) on a single Cloud TPU device.High accuracy, tooFor customers that require the absolute highest image segmentation accuracy, the ShapeMask model can be trained to an accuracy of 38 mask mAP on a Cloud TPU v3 device with a batch size of 64. For comparison, our reference implementation of Mask R-CNN trains to an accuracy of 37.3 mask mAP on a Cloud TPU v3 device with a batch size of 64 and requires about 200 more minutes than ShapeMask to train.All the statistics mentioned above were collected using TensorFlow version 1.14. While we expect you to get similar results, your results may vary. More information about ShapeMask and its performance is available in this methodology doc.Getting started—in your browser or in a Google Cloud projectIt’s easy to start experimenting with ShapeMask right in your browser using a free Cloud TPU in Colab. (Just select the “Runtime” tab and “Change Runtime Type” to “Python 3” and “TPU”).You can also get started with a well-optimized, open-source implementation of ShapeMask in your own Google Cloud projects by following this ShapeMask tutorial.If you’re new to Cloud TPUs, you can get familiar with them by following our quickstart guide. Cloud TPU Pods are currently available in beta, and we encourage you to contact a Google Cloud Sales representative to evaluate them.AcknowledgementsThanks to everyone who contributed to this post and who helped develop this new capability, including Tsung-Yi Lin, Anelia Angelova, Pengchong Jin, Zak Stone, Wes Wahlin, Adam Kerin, Vishal Mishra, David Shevitz, and Allen Wang.
Quelle: Google Cloud Platform

Get the word out: AutoML Translation goes GA, plus updates to Translation API

Translation is a critical function for many industries, whether it’s a media organization delivering the news or a software company making sure its documentation is understood in many different languages. As a result, many businesses are using machine learning to help them translate faster and more efficiently than ever. Today, we’re excited to announce updates to our Cloud AI translation services, including the general availability of AutoML Translation and Advanced features on Translation API.How customers are using AutoML Translation Since announcing AutoML Translation in beta at Next 2018, we have seen customers use the technology to build domain specific translation into their workflows. One such example is Bloomberg, who is using it to provide their analysts with a competitive edge in a fast moving industry. “It’s extremely important for our customers to access information from wherever they are,” says Ted Merz, Global Head of News Product at Bloomberg LP. “We decided we wanted to add the ability for people across the world to see news translated into the language that was most useful to them. Google’s AutoML Translation helped us improve the fluency of translation for financial jargon and terms, ensuring that the stories on Bloomberg’s First Word service could be delivered in real time.” Updates to our Translation APIAlongside the general availability of AutoML Translation, we’re also expanding the ways customers can use our Translation API, by introducing Advanced and Basic editions of the API. By offering two versions, we’re giving customers the flexibility to choose the functionality that best serves the needs of their business.The first edition, Translation API Advanced, includes our latest feature updates to Translation API. We will continue to build in advanced features to this API, such as Glossary (see image below) and model selection which were launched this year. This edition of the API is an iteration of our Translation API v3 that launched at Google Cloud Next and is now generally available.Use the Glossary to translate brand and domain-specific terms with Translation API AdvancedTranslation API Advanced will only support Service Accounts for authentication, so that customer managed resources can be protected with secure authentication. For a more detailed walk through of developing with Translation API Advanced, check out this blog post.One customer that has been evaluating translation APIs is AllTrails, who helps people explore the outdoors with the largest collection of detailed, hand-curated trail maps as well as trail reviews and photos crowdsourced from a community of over 10 million users. Now that the Translation API Advanced is generally available, they’re excited to put it into production in their organization. Says James Graham, Head of Engineering at AllTrails, “Landmark and hiking trail names can have many possible translations that could be correct in different contexts, and having inaccurate translations can result in a bad day on the trail. We are looking forward to using the Glossary feature on Translation API Advanced to enable us to control the names of places in the translation results provided by Google’s great machine translations.”The second edition, Translation API Basic, is designed for customers integrating Translation services into chat applications, social media and gaming, and makes our API as easy to use as possible. Translation API Basic, the new name for Translation API v2, includes support for API keys, and the simplicity and quality the API is known for, and we’ll continue to deliver quality improvements to Translation API to support our customers.With this announcement, Translation API Advanced joins Translation API Basic as a GA service. You can learn more about our Cloud Translation services on our website. For a more technical review of the two editions of the Translation API, view the documentation.
Quelle: Google Cloud Platform