CloudCore Overview

Core Concepts

The Cogniac CloudCore system makes it insanely easy for anyone to harness deep-learning powered computer vision techniques to classify, detect, count, locate, or measure items, actions, or conditions of interest in images or video.

The core concepts of the Cogniac system are as follows:

Media
Media corresponds to still images or video. Media can be uploaded into the Cogniac system via our API, our web application, our mobile application, or our edgeflow gateway systems. The Cogniac system can also actively acquire media from external entities such as network cameras, machine vision cameras, or via other standard protocol sources.

Applications
All visual processing work performed by the Cogniac system occurs in the context of an application. An application performs a single visual task to identify, locate, or report important information within image or video media. The Cogniac system supports a number of different types of applications for working with individual media items or groups of related media items. Many of the Cogniac application types are based upon, and encapsulate, deep convolutional neural network models that perform classification, detection, measurement, localization, or counting of objects, conditions, and actions of interest within images or video. At the core of the Cogniac system applications work by processing input media and associating that input media with one or more user-defined concepts known as ‘subjects’. In the simplest form applications label, or tag, images with subjects that are associated with the image.

The Cogniac system supports advanced application pipelines where output media of one application is routed into the input of another application. In this fashion complex processing tasks can be composed from simpler operations.

Subjects
A subject is the central means by which media is grouped, managed, and routed within the Cogniac system. Subjects are arbitrary concepts that are defined by the users of the system and associated with the visual media. Subjects are generally related to the goals of the visual observation task that is being automated. In the simplest form a subject can be thought of as a tag that can be visually associated with an image. In the fullest form a subject represents any user-defined concept that can be present in, or associated with, the visual content or domain of a group of images or video. Subjects can be simple concepts such as “cat” or can represent more complex concepts such as “cat with mouse in mouth”. A subject can also be the “domain” of an image generation process, such as “cat door camera”, or any such logical grouping of related images. In many real-world cases it is common for a subject to be related to an undesirable condition such as a manufacturing defect that may be present in an image.

Subjects are used to specify both the input of applications (e.g. the set of images an application should process) and the output of applications (the concepts that the application is classifying, detecting, counting, measuring, etc). In advanced use cases the output subject of one application can be the input subject of a subsequent application, enabling sophisticated processing pipelines.

Data Flow

Media flows through the Cogniac system via subjects and applications.

The below diagram shows the flow of media data through a single application. Users first upload media (images or video) into an input subject, for instance 'gear images' in the example below . So-called input subjects feed media into an application which performs a specific visual task, such as detecting if the part in the image contains a defect. The application processes this input media using a trained neural network model, which results in the media and the associated prediction being associated with the output subject. In this example the output subject is 'defective gears'. Some of these application outputs (predictions) are surfaced to the user for confirmation and become consensus items used for training purposes.

1708

For more information regarding specific application workflows, see Cogniac Applications

Data Model

Cogniac follows industry best-practice security standards through a private tenant architecture model, ensuring that your data is siloed and protected. Only users that you invite to your tenant will have access to your media, subjects, applications, and associated data.

2240

Quick Setup

Set-Up
User creates a Tenant, and invites other users to join his or her tenant. User then creates their first Application in that Tenant.

Getting Started: Applications
Getting Started: Subjects

Application Training
User adds media to their application via an input subject, and captures that media into their application. Then they provide feedback on that media, training the application to automatically detect, measure, or count the items of interest.

Detect Items or Conditions of Interest
As the user uploads and captures media into their Input Subjects, the Application begins to generate detections. These detections are automated or human-generated detections that are created from items of interest within the users visual media.