Mindtech Chameleon Packs

Chameleon's modular Packs  provide market and application specific environment, data and features.

Overview

Chameleon has been deliberately designed in a modular, expandable way, separating the core functionality from the application specific requirements.

Mindtech offer Application packs, Asset Packs, Scenario and Sensor Packs, to enable our customers to rapidly create the training data they need.


Asset Packs

The asset packs bundle together a number of similar elements.  These could be generic, applicable to many different scenarios such as a variety of humans dressed in casual clothing; or they could be specialised such as a variety of human assets dressed in uniforms appropriate to a retail environment.

 

The assets in a pack may be animate (such as vehicles, humans, animals), or inanimate objects such as items of food, computers and so on.

 



Application Packs

The application packs bring together a scene, assets and behaviours to create a complete environment optimised for a specific market.

Mindtech Application Packs are tailored to individual applications, to ensure the content of the packs allows customers to rapidly create the training data they need to achieve the required levels of accuracy, whilst avoiding undesirable bias.

The base scene is carefully chosen to allow for the use of Domain matching techniques, using Mindtech's curation manager to further augment the dataset.

Scenario Packs

One of the key elements of training any neural network is having the right data.  If we have several million images of the normal cases, the system will behave as expected in the majority of cases.  However in those difficult to understand and obtain data for corner cases, the results will not be as desired.  If we have hundreds of images of white cars, then we risk only understanding white cars, or worse still confusing any white object as a car.

 

Mindtech scenario packs utilise the expertise of our machine learning engineers, working in conjunction with our customers to create the ideal set of conditions to ensure that all the required cases are covered by your dataset.  The scenario packs include the scripts to run the simulations at the relevant times of day, with the assets, cameras in the correct locations

 

Sensor Packs

For any training data to give the best results, it is vital to model the inference domain as accurately as possible.  That is to say, we need to model the deployment conditions that the system will be exposed to; this extends beyond the scene look and feel, right to the capture system itself, including the characteristics of the sensor system.  

For a camera, the sensor system has a large number of elements, all of which will introduce distortions to the captured image, from lens aberrations and geometric distortions, through sensor noise and coloration to compression artefacts caused by the ISP.

Mindtech sensor packs can create data that has been "pre-distorted" to match the deployment domain, leading to improved accuracy.

Chameleon AI Tools & Services

The Mindtech AI tools and services family comprises a comprehensive set of tools designed to create, manage and generate synthetic data for neural network training.