On August 15, 2022, Nearmap released Generation 5 (Gen 5) of Nearmap AI. All of the data layers in Nearmap AI have been updated and enhanced by the more powerful and accurate AI and machine learning algorithms used to develop the Gen 5 data layers. Gen 5 more accurately identifies features in our imagery.
What does this mean for my models built for Gen 4 data layers?
Gen 5 uses a new deep learning model to power all the raster and vector layers. Trained on more than 2.5 times more data than Gen 4, Gen 5 improvements have solidified earlier AI results while improving others. For many feature classes, specific edge cases have been improved, while keeping the same class definition.
Due to the wide variety of use cases and applications, Nearmap recommends validating models built for Gen 4 against Gen 5 data before deployment.
To support validation, Nearmap is running Gen 4 and Gen 5 models on all standard vertical imagery captured between July 1 and August 15.
How do I specify AI Generations in API Calls?
After August 15, Generation 5 will be the default response from the API for any surveys where Gen 5 was processed. A special query parameter can be included in API calls to return Gen 4 data. Setting the API flag to “Gen 4” will return Gen 4 data. No setting or a “Gen 5” setting for the API flag will return Gen 5 data. You can refer to the documentation on the AI Feature API for more information.
Again, Gen 4 results will only be available for surveys captured before August 15. Gen 5 results will only be available for surveys captured after July 1.
What steps are recommended for deploying existing models for Gen 5 data?
All models and algorithms are unique and we recommend following the best validation process that works best for your models and your organization.
1. Lock out Gen 5
Adjust the API calls in your models to use the “Gen 4” API flag to ensure that no Gen 5 data is returned from the API.
2. Update code to pull Gen 5 data
Use your production systems to pull Gen 5 data via the API, by using the “Gen 5” API flag. While the default setting will pull Gen 5 data by default, in areas without Gen 5 data, it will return Gen 4 data. By setting the “Gen 5” API flag, you are ensuring that data returned is Gen 5 data.
3. Validate model results
Test the performance of your models using both Gen 5 and Gen 4 data at the same locations and dates. Examine subsets of data where the different generations deliver different results.
4. Retrain the model as needed for Gen 5 data
It is likely that some benefit can be obtained through retraining your models on Gen 5 data, even if it behave appropriately on Gen 5 data during validation.
Will the Gen 5 model be run on imagery captured prior to July 1?
We are reprocessing some surveys captured prior to July 1, 2022. We anticipated more than 1 million square kilometers of historical imagery will also be reprocessed for Gen 5 data. This data will be available by the end of September 2022.
Where can I find more information about Gen 5 AI content?
Visit the Nearmap Generation 5 AI Content for more information about Gen 5.
You can also reach out to your account representative (CSM) to learn more about how Nearmap AI Generation 5 can help improve your workflows.
Sr. Product Marketing Manager for Insurance based in Charlotte, NC
Richard is the Sr. Product Marketing Manager for Global Insurance at Nearmap, Inc. He has more than 14 years of experience developing strategy, positioning, and messaging for desktop, enterprise, and SaaS software and technology, including imagery, data, and open source solutions.
Prior to joining Nearmap, Richard worked at Deloitte, where he helped bring to market Close As You Go, a cloud-based software platform that used blockchain technology to help state and local governments manage disaster recovery documentation to demonstrate eligibility for FEMA Public Assistance expenditure reimbursements. He has also worked at EagleView, Esri, and the USGS. He was a Featured Speaker at the 2016 Esri Public Sector CIO Summit and two President of Esri Toastmappers public speaking club.