Project for the course "Visual Information Processing and Management" at University of Milan Bicocca (a.a. 2021/2022).
The dataset used as training set is Annotated Potholes Image Dataset.
The detection task is carried out with YOLOv4 and YOLOv4-tiny, implemented in the Darknet framework.
The datasets, the PowerPoint presentation and the best models' weights are available here (unfortunately, the presentation is only avilable in italian).
Main pipeline:
Some results on images:
Some results on videos (the gif takes a few seconds to load):
