Citadel AI is announcing the release of Citadel Lens for Object Detection, our product for automatically testing and validating object detection AI models.
In addition to classification and regression AI models (on tabular and image datasets), Lens now supports object detection models across industries such as medical devices and autonomous driving. With our unique model-agnostic testing technology, Lens can easily connect to existing AI systems without code modifications.
International regulations and standards on AI reliability
The European Union is finalizing a landmark regulation on the usage and trustworthiness of AI systems, known as the EU AI Act. Similar to GDPR, this regulation holds companies accountable with fines reaching up to €30 million or 6% of global sales for non-compliance. In addition, ISO, IEEE, and other international organizations are actively developing technical standards for measuring the robustness, fairness, and explainability of AI systems.
Object detection systems are critical in industries such as healthcare, autonomous driving, and product inspection. In these domains, models need to accurately detect and recognize multiple objects in a range of environments.
However, data quality can vary greatly across real-world environments due to discrepancies in the camera hardware/software, background environment, or positioning. This causes low AI performance during training and deployment, and it can be extremely time-consuming to validate that the AI system performs reliably across numerous edge cases.
Poorly tested AI systems can result in not only technical bugs and financial losses, but also pose risks to human safety, brand reputation, and even non-compliance with emerging regulations and standards. Citadel AI provides a solution to these AI reliability risks.
Automatically explain, test, and debug object detection models
Common problems when testing object detection AI models include bounding box misalignment, missed detections, false detections, and classification errors. Citadel Lens for Object Detection automatically visualizes, detects, and quantifies these problems.
Explaining object detection models is often difficult since there can be multiple predictions per image. Citadel Lens allows you to easily connect to an existing model and interactively explain all model predictions in an image, visualized as pixel heatmaps.
Additionally, Citadel Lens automatically generates synthetic test data based on real-world environmental conditions, such as changes in camera hardware, changes in the environment, and changes to the position of objects, allowing customers to immediately identify and improve areas of low model robustness.
The Lens Bias Finder automatically discovers weak areas of the model – slices of data where a model has abnormally low performance. This makes it easy to slice AI performance across dimensions such as the location of the image, the data annotator, the brightness of the environment, the focus of the camera, and so on.
Citadel Lens can be applied to a wide range of AI applications, from tabular applications such as credit scoring systems, demand forecasting, and logistics optimization, to image applications such as medical imaging, manufacturing visual inspection, and content moderation.
By automating QA and testing of model reliability, Citadel Lens helps protect you from business risks, technical risks, and compliance risks.
About Citadel AI
Citadel AI provides automated testing and monitoring products for AI applications, to help organizations minimize AI reliability risks, and maximize AI performance from research to deployment.
Citadel AI’s co-founder was the former product manager for ML infrastructure at Google Brain, and is currently leading Citadel AI’s product and engineering teams.
