Implementation of detection by transfer learning with popular Yolo8 and Fast-RCNN from Detectrons zoo. Training takes place on a custom dataset assembled from several.
The dataset includes 157 classes:
- 156 - different types of Russian traffic signs
- 1 - potholes
The repository contains a script for assembling datasets and bringing them to required annotation format.
Both datasets are available at the following links:
Download weights for trained Yolo8m from here.
Not bad!
