Join our community of Clarifai users. Explore and share models, workflows and more.
We're on a mission to deliver "The World's AI." Clarifai Release 7.7 introduces advanced new tools for data training datasets, visual segmentation and AI on the edge. Plus, we're introducing a new model that we hope can make a positive social impact through smart content moderation. Read more to learn about the latest developments in machine learning technology that are giving customers unprecedented insights into unstructured image, video, text and audio data.
The new hate symbol detector can detect the presence and location of ADL-recognized hate symbols in any image. The taxonomy for v1 of our hate symbol model contains two concepts, “confederate-battle-flag” and “swastika.” These two symbols have been chosen as a starting point for v1 because of the readily available training data. We are proud to be able to provide content moderation solutions that can help prevent the proliferation of these hate symbols. Learn more about the model here.
Images can be processed by computer vision algorithms at varying levels of precision. You can process an entire image at once with a classification model, you can process a rectangular region of an image with a detection model or you can process the individual pixels that represent an object with a semantic segmentation model. We're adding support for image masks to our platform so that you can label each individual pixel that belongs to a given concept for use in semantic segmentation. Learn more.
We've listened to how customers are using our Scribe tool to label their datasets, and we're adding functionality that allows users to add or remove workers from active labeling tasks. Workers can only label inputs that have been assigned to them, and special redundancy checks are in place to prevent new workers and existing workers from labeling the same image twice. Try it yourself.
The new Quick-mask tool makes it incredibly easy to select objects with pixel-level precision. Users simply draw a bounding box around an object to have an image-mask returned as a labeling suggestion. These pixel-level selections can be used to create image-masks and train new semantic segmentation. Try it yourself.
In your worker management dashboard for Scribe, task reporter provides an at-a-glance view of team performance and productivity. This new feature helps you ensure that the human component of your data pipeline has the support and guidance that they need to perform their jobs successfully. Metrics like total inputs and task annotations can be tracked over time. Try it yourself.