A geometric and deep learning reproducible pipeline for monitoring floating anthropogenic debris in urban rivers using in situ cameras
Researchers have developed a new pipeline using deep learning and geometric modeling to monitor floating debris in urban rivers. This system quanties debris continuously and identifies the most accurate and efficient deep learning models for complex environmental conditions. The study also demonstrates the feasibility of estimating object size from 2D images and highlights the importance of dataset composition, including negative images and avoiding temporal leakage. AI