Gear Up Your AI: Fine-Tuning LLMs
March 29, 2021

Clarifai Release 7.3

Table of Contents:

The Latest Tech in Machine Learning AI

Clarifai Release 7.3 pushes machine learning farther by expanding upon our ecosystem of AI tools and technologies. We are introducing entirely new model types, as well as augmentations and improvements to some of our most important workflows. We are also taking data labeling to the next level with advanced new Scribe labeling tools



Join our Slack Community

Interact with other people actively using and working with Clarifai products.

Join Slack Group

Clarifai User Slack Community

Clarifai Armada Predict

Logo Detector

The Logo Detector automatically identifies and classifies logos for some of the world's most popular brands. Detecting product logos within images and videos is one of the most popular uses of computer vision AI, as brands look for better ways to protect their identity, monitor usage trends and provide targeted communications to their consumers. Learn more in our model gallery or try the new logo model yourself.


Weapons Detector

The Weapon Detector automatically identifies and classifies weapons in images and videos. AI-powered weapon detection promises to be one of the most important public-safety technologies in production today, and Clarifai has built this model from the ground-up to identify dangerous weapons so that you can deploy your solution quickly.  Learn more and try it yourself. 


General-Text-Token-Classifier (Initial Release)

Tokenization of text passages helps split text passages into small, meaningful chunks. Tokens can be individual words, or even longer passages of text. In this initial release of our text token classifier, our model will automatically recognize common entities that would be present in form documents. Learn more, try it out yourself, and stay tuned for upcoming developments to this model.


Clarifai Mesh Workflows

Face V3 (Landmarks + Alignment)

The new face detection model from Clarifai takes face detection to the next level by combining the latest image recognition algorithms with Landmark Detection and Alignment techniques, so you can accurately recognize faces from any angle. Our Face v3 workflow has been trained on a dataset of 10 million face images with nearly 100,000 accurately-labeled identities. Face v3 offers discriminative features for face recognition that consistently outperform the state-of-the-art. Learn more and try it yourself


Object Character Recognition (OCR) - Scene

Text frequently appears in images and videos and you can now detect text for an international, multilingual user base. Our Object Character Recognition (OCR) - Scene workflow now supports 18 languages. Use images and videos as inputs into your workflows, and receive machine encoded text as an output. Learn more and try it yourself

Object Character Recognition (OCR) - Document

The Object Character Recognition (OCR) - Document Workflow is one of the most powerful tools that you have at your disposal when it comes to understanding text in documents. Now Object Character Recognition (OCR) - Document is more powerful than ever with support for over 162 languages. We've even configured our text-extraction operator to encode a visual inputs containing non-Latin characters. Learn more and try it yourself. 

Scribe Label

Task-Labeler: Polygon-Interpolation

Annotating video presents a unique challenge and a unique opportunity when creating training data sets. By using our interpolation technology you can select an object in one frame of video, select it again in a later frame of video, and then automatically label all of the frames in-between. Interpolation now supports both bounding box and polygon labeling types, so that you can label video accurately and efficiently. Find out more.


Task-Labeler: Timeline Panel for Video Tracks

Our new timeline panel makes it easier than ever to work with video in your labeling tasks. The timeline panel helps you interact with your videos and records keyframes to let you know where you have made adjustments to your interpolation paths. Learn more.


Task-Reviewer: Grid Display

In the new and improved reviewer interface, you can view multiple labeled inputs at once, and accept or reject labels individually or in bulk. You can also toggle back and forth between the labels created by your individual labelers. Find out more.



Clarifai Computer Vision Portal

Portal Data Mode: Improved Collectors with Edit Functionality

Collectors are now easier to use than ever with new edit functionality that allows you to make adjustments to your collectors after they have been configured and deployed. Learn more.