Training Data Generation

Build segmentation data from annotated driving images, public datasets, and carefully reviewed augmentation. The lesson also traces a missed-pedestrian failure back to candidate training examples, showing how targeted collection differs from simply adding more data.

Generating training examples

Start with the conditions in which the system is intended to operate. Collect and evaluate segmentation data across those conditions, including failures and uncommon scenes. A finite training set cannot contain every possible real-world event, so coverage claims need a defined scope.

Human-labeled data

First, you will have to gather driving images and hire people to label the images in a pixel-wise manner. There are many tools available such as Label Box that can facilitate the pixel-wise annotation of images.