Možnosti platformy AI Vitis pro řešení úloh umělé inteligence v biomedicíně

Abstract

This scientific report focuses on the KriaKV260 development board designed by Xilinx and its application in implementing an artificial intelligence model for a biomedical purpose. In addition, the report investigates the pre-built applications that KriaKV260 suggests and documents how to add and change the models of these pre-built applications. The study highlights the various AI tools used in the investigation, including the Model Zoo, the Inspector, the Quantizer, the Compiler, and others. The report provides insight into the different capabilities of KriaKV260 and the tools available for AI scientists to create and deploy AI models for various biomedical applications. Specifically, an AI model was developed to detect Pneumonia in X-ray images, with training performed on GPU and optimization techniques employed using Vitis AI tools to enable deployment on Kria. However, the report also notes the challenges involved in using Kria for custom models and applications, requiring additional tools and development platform knowledge beyond pre-built or pre-trained models.