Accelerating AI development with thermal imaging.

The data lifecycle management (DLM) application optimizes dataset development with access to an annotated infrared and visible library for neural network training and advanced model performance testing.

Conservator™, a cloud-based dataset development subscription software for perception engineers using thermal infrared and visible image datasets to train neural networks has been announced by Teledyne FLIR.

Subscribers also gain access to application specific segments of their annotated image library of more than one million images with more than 100 object label categories. Designed to meet the workflow demands of data scientists in automotive, security, and smart cities applications, Conservator scales to support enterprise artificial intelligence (AI) teams in the research and development of object detection models.

“Conservator is a powerful application for data scientists developing datasets with a full complement of workflow functions including annotation, version control, data right access and model performance,” said Arthur Stout, director of AI product management at Teledyne FLIR Infrared Imaging OEM. “AI starts with quality data and this application supports collaboration to advance multi-sensor neural network development in commercial and defense AI applications.”

Conservator includes dataset workflow tools for annotation, curation, quality assurance, and dataset version control. Built on a scalable and stable database, Conservator can manage petabyte scale libraries. In addition, the included Conservator Insights™ desktop tool provides analysis and visualization of model performance against ground truth references. This empowers data scientists to quickly pinpoint the specific images in large datasets causing false positives or missed detections, enabling rapid dataset iteration and neural network re-training.

Conservator follows the recent release of the Teledyne FLIR expanded free starter thermal dataset for Advanced Driver Assistance Systems (ADAS) and self-driving vehicle researchers and developers. With both thermal and visible annotated images across fifteen object categories, the free starter thermal dataset allows the automotive and academic community to quickly evaluate the vehicle safety algorithm performance, neural network testing, and thermal sensors, such as the FLIR ADK™.

@flir @mepaxIntPR #PAuto #Automotive