CLARIFIER - Data Labeling and Curation at Scale (DLCS) for Machine Learning Algorithms

The CLARIFIER system is designed to revolutionize data labeling for complex 3D CT scans and mm-wave radar data, addressing key challenges in security and healthcare. In Phase II, CLARIFIER will extend to handle 3D volumetric data, focusing on automatic labeling, efficient human verification, and interactive correction. The system will use advanced techniques, such as tiered hardness-based labeling, foundation models like SAM (Segment Anything Model), and consensus-based methods to improve the labeling process. It will also feature data filtering and targeted selection to enhance data quality and relevance. In Phase I, CLARIFIER achieved 5x to 10x speedups in labeling time and cost, requiring 30-40% fewer samples. The goal is to develop a near-perfect model orders of magnitude faster than current labeling solutions. A user-friendly web and mobile interface will enable users to visualize and correct labels slice-by-slice, making it accessible for both experts and non-experts. The system will also include a robust data ingestion pipeline supporting formats like DICOM, DICOS, HDF5, and TIFF, ensuring smooth integration with DHS data systems. The final system will be thoroughly evaluated through human studies, showing significant improvements in labeling accuracy, speed, and efficiency. This scalable solution will have broad commercial applications in areas like baggage screening, autonomous driving, and medical imaging, providing powerful support for DHS and beyond.