We will share our experience and lessons learned in choosing, training and deploying deep learning models for single urban tree detection, localization and classification using street-level imagery. Drawing on our case study of Canada-wide urban tree mapping, we provide guidance on questions such as: what to think about when choosing a Deep Learning framework for your computer vision use case? How can you be creative in acquiring your training data? How do you train an instance segmentation model with limited labeled examples? Which open source projects to choose to quickly get started?