September 29, 2021

Clarifai Release 7.9

Clarifai Release 7.9

Clarifai is a data-centric AI platform that the world uses to develop and deploy deep learning models. Our suite of integrated tools makes it possible for business users, data scientists, and developers to design AI-powered solutions in the real world efficiently. In release 7.9 of the Clarifai platform, we introduce powerful new tools for working with video data, datasets, and people detection.




Task-Labeler: Added Time-Segment tool

When working with temporal data like video, it can be helpful to work with video inputs as if they were sequential sets of individual images. This approach allows the users to use tools like video interpolation and tracking to label large numbers of unique images. Tagging images in videos is at least 100x faster than labeling images that are not part of a time sequence. But what about when you need to label an entire time segment as one concept? You can now do this in Scribe Label. Find out more.

Annotation: Added support for Time-Segments

API users are now able to add a Time-Segment annotation to a Video-Input via the PostAnnotations-endpoint. You can create a time segment object and work with new TimeSegment fields in your DataExamples. Get started using Time-Segments via API.


Dataset: Initial Release

A dataset is a set of data examples. With Clarifai datasets, you can manage the datasets you want to use for visual search, training, and evaluation. Up until now, saved searches have been used to generate datasets dynamically. With the new dataset entity, datasets are stored as convenient snapshots in datasets tables. Learn more.

Dataset-Version: Initial Release

Dataset preparation is not a one-off task. You need to systematically change and enhance your datasets to improve the accuracy of your AI system. Dataset versions to are used to help you experiment with and iterate upon your datasets over time. Ensure that you have sufficient data for your problem, consistency in data annotation, representative samples of the data that you expect to see in deployment, and coverage of all variations that deployment data will present. Learn more about dataset versions.


New Model Version: person-detection-yolo (Yolo5x)

Our most accurate and performant person detection model yet. Our new Yolo5x person detector is 6x faster than our current person detector model. Identify people in photo and video data. The people detector returns one class: "people," along with bounding boxes and probability scores. Learn more and try it yourself.