17738754. ADAPTIVE ARTIFICIAL INTELLIGENCE FOR THREE-DIMENSIONAL OBJECT DETECTION USING SYNTHETIC TRAINING DATA simplified abstract (MICROSOFT TECHNOLOGY LICENSING, LLC)

ADAPTIVE ARTIFICIAL INTELLIGENCE FOR THREE-DIMENSIONAL OBJECT DETECTION USING SYNTHETIC TRAINING DATA

Organization Name

MICROSOFT TECHNOLOGY LICENSING, LLC

Inventor(s)

Wolfgang Martin Pauli of Seattle WA (US)

Mario Emil Inchiosa of San Francisco CA (US)

Lingzhi Allen of Kirkland WA (US)

Daniel James Haines of Emmbrook, Berkshire (GB)

Matthew Anthony William Hyde of Brighton (GB)

ADAPTIVE ARTIFICIAL INTELLIGENCE FOR THREE-DIMENSIONAL OBJECT DETECTION USING SYNTHETIC TRAINING DATA - A simplified explanation of the abstract

This abstract first appeared for US patent application 17738754 titled 'ADAPTIVE ARTIFICIAL INTELLIGENCE FOR THREE-DIMENSIONAL OBJECT DETECTION USING SYNTHETIC TRAINING DATA

Simplified Explanation

The patent application describes an adaptive AI model for 3D object detection using synthetic training data. The model is trained to detect specific items of interest by generating a training set in real time during the training process.

• The training set consists of composite images that depict containers packed with items of non-interest and isolated images of the items of interest.
• Multiple images are generated during each training iteration to create a diverse and comprehensive training set.
• The trained AI model can then detect items of interest in actual containers and provide a classification indicating the likelihood of finding such items.

Potential applications of this technology:

• Enhancing object detection in various industries such as logistics, retail, and manufacturing.
• Improving security systems by accurately identifying specific objects or items of interest.
• Streamlining inventory management processes by automating the detection of specific items in containers.

Problems solved by this technology:

• Overcoming the limitations of traditional training methods that rely on manually labeled datasets.
• Addressing the challenge of limited availability or high cost of real-world training data.
• Providing a more efficient and scalable approach to training AI models for object detection.

Benefits of this technology:

• Enables the generation of large-scale, diverse, and realistic training datasets in real time.
• Reduces the reliance on manual labeling and the need for extensive real-world data collection.
• Improves the accuracy and reliability of object detection systems by training on synthetic data.

Original Abstract Submitted

Embodiments described herein are directed to an adaptive AI model for 3D object detection using synthetic training data. For example, an ML model is trained to detect certain items of interest based on a training set that is synthetically generated in real time during the training process. The training set comprises a plurality of images depicting containers that are virtually packed with items of interest. Each image of the training set is a composite of an image comprising a container that is packed with items of non-interest and an image comprising an item of interest scanned in isolation. A plurality of such images is generated during any given training iteration of the ML model. Once trained, the ML model is configured to detect items of interest in actual containers and output a classification indicative of a likelihood that a container comprises an item of interest.