For this project, a thousand images were uploaded and manually tagged as 'good', 'bad', or 'acceptable' as training data. A custom classifier was built on top of this training data and used to classify incoming bike lane imagery as in "good", "bad", or "acceptable" condition.