Anomaly

Introduction

In order to understand which part of an image contains odd things by a computer vision model, traditionally, supervised learning would be required which involves large volumes of annotated examples and hours of training. On EdgeAI Studio (opens new window), we introduce you the "Anomaly" block, a self-supervised visualisation model pre-trained using 14 millions images, which is ready to use without programming. With the "Anomaly" model, you are able to discover and segment any unlabelled objects from an image or a video with absolutely no supervision and without being given a segmentation-targeted objective. It will be able to focus on certain parts of the input automatically which contains odd objects. This kind of segmentation will help facilitate tasks ranging from swapping out the background of any images or videos to teaching AI systems that navigate through a cluttered environment.

Usage

Our Anomaly model automatically learns an interpretable representation and separates the main object from the background clutter. It learns to segment objects without any human-generated annotation or any form of dedicated dense pixel-level loss. Computer vision or AI systems that are far less dependent on labelled data and vast computing resources for training can be built effectively. Object detection model for object prediction, object tracking and/or object counting can be built easily with the help of Anomaly model.

Examples

Prediction and Bounding Boxes creation with Anomaly Block

To use Anomaly block for creating bounding boxes for prediction, you can set up a pipeline as below:

You may find a demo pipeline here (opens new window).