AIM:

• To fit a YOLO12n Segmentation model on the a publicly sourced, but partly modified dataset.

Procedure:

• Use pip install ultralytics to install the prerequisites (venv recommended).
• clone this repo
• use bash or psh and execute python train.py

Expected out:

Demo:

Updates:

• Added a new demo to see the segmentation done on a sidewalk for ISC's welcome day demo.
• Trained a new model to have only one class: "road". This is part of a new direction for the club to improve the perception node.
• Negative slopes are considered as the right contour while the positive ones are left contours. This is implemented in the ROS node for ISC's phoenix repo. (Justification for the red and green edges in the demo video)
• The earlier model looked like this:[]