Hardware improvements coupled with advancements in frameworks, data and algorithms has produced the means to classify our own images. One of the essential aspects of this technology is the detection of objects within images.
Whilst accessible, it remains difficult to develop a solution to address a specific business problem. AWS has made it easier to build and deploy machine learning models. Amazon Rekognition can be used to provide accurate facial recognition and has the facility to compare and analyse faces. Adding these features to any application is simple if the Amazon Rekognition documentation is followed.
Faster Model BuildingFor custom image classification, a platform such as Amazon SageMaker is required. This works all the way through the machine learning model’s development. Choose from a number of built-in algorithms to reduce time and cost. The image classification algorithm is an integral part of the system and prepared for transfer.
The aim is to accelerate the process of incorporating machine learning into new applications. An easy way to train and deploy models for developers, it is a fully managed framework that enables building, training and hosting models at scale.
Three-Stage Build ProcessThe tool comprises three sections starting with a notebook using Jupyter notebooks for the review of data constituting the basis of the model. This initial stage can be run on standard instances or GPUs for more resource-hungry processes.
When the data is ready, the model must be trained. This step introduces the base algorithm to be used for the model. AWS has included pre-configured algorithms within the tool. Every built-in algorithm within Amazon SageMaker is a Docker image, already prepared with binaries, frameworks and libraries. However, it is possible to design algorithms from scratch by generating a Docker image. This flexibility represents a feature of the system. The service can deal with most popular algorithms whether bespoke or built-in.
Amazon looks after all of the supporting infrastructure needed to operate the model, including problems like auto scaling, node failure and security patching. Once the model has been created, it can be run from SageMaker. The service is available for free up to certain usage levels as part of AWS free tier of services. Pricing based on region and usage once the threshold is exceeded.
Data SecurityAll API and console requests are made over an SSL connection with maximum security and model and system artefacts are encrypted in transit and when at rest. Additional encryption can be applied to data by the use of encrypted S3 buckets and a KMS key to training jobs, notebooks and endpoints.
Amazon SageMaker helps to focus on the creation process to provide an effective tool for image-classification and a sturdy support for the entire production pipeline.
Maybe you'll find this webinar interesting: