News & Events

Enhancing AI at IREX: The Strategic Role of Synthetic Datasets in Training Neural Networks

Demo Corporate
IREX, a leader in ethical AI solutions for public safety, has achieved significant advancements in its AI-driven video analytics capabilities through the integration of synthetic datasets into its training pipelines. This strategic innovation not only enhances the precision of our systems but also supports our commitment to upholding human rights and ethical standards in surveillance technologies.

Enhanced Detection Capabilities with Synthetic Data

Recent enhancements in IREX’s video analytics have notably improved our ability to detect critical incidents across public spaces such as railways. By incorporating 5,000 synthetic images into our finetuning set, complementing 6,000 real-world captures from our extensive camera network, we've seen a measurable increase in model accuracy. The mean Average Precision (mAP) of our detection algorithms rose by 4.6%, a testament to the efficacy of blending synthetic and real-world data in training sophisticated AI models. The pre-train dataset consisted of 50,000 real-world images.

Addressing the Challenge of Data Scarcity

One of the primary challenges in developing advanced behavioral analytics for public security applications is the scarcity of relevant training data. Events such as incidents in airports, highways, and subways are relatively rare, making it difficult to gather the extensive datasets required for machine learning. To overcome this, IREX has pioneered a hybrid data acquisition strategy that combines authentic footage with synthetically generated content, ensuring both variety and volume in our training datasets.

Innovative Production of Synthetic Data

Our approach to synthetic data generation involves close collaboration between machine learning scientists, 3D artists, and generative AI specialists. This team effort leads to the creation of highly randomized and diverse scenarios that significantly enrich our training libraries. The process is not only innovative but also cost-effective; producing 50,000 synthetic images of subway incidents, for example, requires significantly fewer resources compared to the extensive effort needed to capture and label equivalent real-world data.

Strategic Focus Areas

Currently, IREX is focusing on expanding its synthetic data generation to cover key areas of public safety concern, including detection of weapons, people falling or lying down, and distinctive car features. This targeted approach allows us to streamline our development cycles and significantly boost the performance metrics of our video analytics systems.
By strategically incorporating synthetic datasets into our AI training regimen, IREX is not only advancing the frontiers of public safety technology but also ensuring that these innovations are developed in an ethical, efficient, and highly effective manner. As we continue to enhance our capabilities, our partnerships, such as with the National Child Protection Task Force, are invaluable in our ongoing efforts to safeguard communities while respecting individual rights and freedoms.