October 30, 2020

Clarifai Release 6.9

Table of Contents:

With Clarifai Release 6.9 we are introducing a ton of major platform improvements and new features that help you get the most out of AI in  your business. Learn about ingesting and labeling new datasets, new models and model-training capabilities, and more.

AI Assist and Auto Annotation Improvements

aiassist

With Scribe, machine learning is at the core of your data labeling pipeline. You can label data automatically, or send it to your (or our) human workforce for review. This gives you the power to tailor your AI solution to your business needs, and prevent "data drift" by keeping your models freshly trained with production data. 

Collector Improvements

collectorimpr

We've made collectors easier to use and more powerful than ever. Collectors help you collect inputs into your app by adding a new input every time a model is called on this input. This means that you can automatically collect training data as new predictions are made, so that you can keep your dataset up-to-date. 

Deep Training Videos in Scribe

deeptrainvid

You can now deep train with video in Scribe. Video is information-dense and one video can provide the equivalent of thousands of individual image inputs to your training dataset. With a dedicated UI and intuitive tools, it has never been easier to build highly accurate and specialized deep-trained models.

Consensus Review in Scribe

consensusrev

Consensus review helps you annotate data at scale by automatically leveraging the power of worker consensus in your labeling tasks. Consensus review can automatically detect and compare similar annotations provided by your labelers, so that you can automatically approve labels where people agree on what they are seeing. This saves you  significant amounts of time in when labeling data, and helps ensure quality at scale. 

People and Vehicle V2

peoplevehiclezoom

Real-world applications of object detection are run on a variety of platforms, which often demand different resources. Our new People and Vehicle Model V2 uses efficientdet d2 to deliver better accuracy, with fewer computational resources than our previous architecture. The new model is more than 6x faster compared to our current People and Vehicle detector model. 

API Improvements

armada

Support for the secure gRPC channel has been released, along with many other behind-the-scenes improvements to the way we support our API. Binary gRPC channels are faster - demanding ~30% less traffic on the network. It is also easier to generate new clients. We've also added HTTP/2 support for our REST client interfaces, a single domain for our API, and improved routing management. 

App Pinning

apppinning

Easily keep track of your most important apps by "pinning" them in the app overview page.