D5.3 : Demonstration of the Tool for Industry Application Case and KPI Measures

Abstract

This deliverable presents the final demonstration and validation of the tools and AI models developed for the EXA4MIND Industry Application Case, specifically targeting Advanced Driving Assistance Systems (ADAS). It outlines a data processing pipeline that leverages the Advanced Query and Indexing System (AQIS) for scalable, AI-based pre-annotation and smart data retrieval on extreme-scale datasets. The report validates the successful achievement of three primary Key Performance Indicators (KPIs). For KPI 6.1, the deployment of models such as YOLO v11, YOLO World, and SAM resulted in a 76% reduction in manual annotation time for object detection and a 74% reduction for semantic and instance segmentation. For KPI 6.2, the project built a robust database encompassing 19,259 hours of multi-sensor driving data. Finally, to fulfill KPI 6.3, the document demonstrates a ”smart database querying” mechanism that fuses natural language and image similarity searches to rapidly retrieve and analyze ADAS failure scenarios. These advancements enable targeted data curation and rapid, closed-loop AI adaptation, significantly improving the efficiency of ADAS development