Nearmap, a specialist in the field of aerial imagery, has announced the impending release of its sixth-generation AI model that has been developed utilising AWS. The company has reduced the training time required for its AI model from up to eight-weeks to just 26 hours, according to a statement. This approach to training represents a 98% decrease while also benefitted from more than forty times the compute power.

This use of artificial intelligence (AI) and machine learning (ML) has allowed Nearmap to offer automated processes and accuracy typically unachievable through human input alone. Nearmap is recognised as the only aerial intelligence company that utilises an AI system routinely on every survey flight. The result is the generation of automated high-resolution maps of AI and ML content, covering nearly the complete Australian population.

Unlike satellite companies that capture global images, Nearmap reportedly provides a more detailed solution. With cameras mounted on aircraft, Nearmap can deliver high-resolution imagery of cities, suburbs, and individual properties. This precision provides Nearmap with the ability to offer actionable insights such as identifying surface imperfections, managing inventory assets, and even allowing precise measurements of sites and buildings.

The use of this detailed AI data has proven useful for organisations such as local councils, who have used the data to map vegetation coverage leading to more community-focused decision making. Furthermore, the insurance industry is being transformed by Nearmap's AI capabilities, as insurers can now make underwriting and claim decisions with greater accuracy and sped.

Nearmap's Vice President of AI and Computer Vision, Michael Bewley faced multiple challenges when his team began applying machine learning on over 50 petabytes of imagery data. In his words, "From an aerial perspective, no-one has an AI system that runs at high resolution, routinely, on every survey flight to build automated maps - except us."

Near map has made significant strides since 2019, when it spent about $1 million on AWS compute and produced deep-learning results in Australia and the U.S. The first-generation of AI at Nearmap included 12 different layers in a single model, including swimming pools, buildings, trees and construction sites. This success motivated Bewley to ask his team "to figure out how to put all the layers in a single model". He stated, "We knew the benefits would be huge, so our mantra became: build for a thousand layers."

However, AI training was not without cost. Nearmap had amassed around 1 million labelled images, with a total catalogue of more than 50 petabytes of data. The models were being trained in a data centre taking between six to eight weeks to train. To address this, Nearmap took a bold step by rewriting its training stack, switching from on-premise data centres to AWS. Therein, it harnessed the use of AWS SageMaker, and increased the size of its model architecture and its labelled dataset.

Sarah Bassett, Head of Software, AWS Australia & New Zealand, acknowledges the importance of this efficiency, stating, "The training of AI models can be costly and time-consuming. If companies can gain some efficiency, it will help them accelerate the introduction of new features to the market."

Reflecting on Nearmap's journey, Michael Bewley highlights how shifting to cloud computing on AWS-SageMaker saved valuable time in the model training process while enhancing performance. This transition also unlocked the potential to employ hundreds of GPUs at once whilst bypassing concerns about infrastructure and hardware management.

With the recent unveiling of Nearmap's newest camera system, HyperCamera 3 (HC3), Nearmap continues to push the boundaries of what is possible with AI and ML, reimagining machine learning capabilities from an aerial perspective.