I have now reviewed the manuscript entitled “DeepForest: A Python package for RGB deep learning tree crown delineation” and I can safely say that it was one of the most enjoyable manuscripts I have reviewed this past year. This package represents a tremendous amount of work and I strongly support its publication. Both the main text and the original python code are well written and clear to follow. I believe it will be a critically important example of the application of deep learning tools for biological questions. I have only minor comments.

First and most importantly, the basic usage demo does not work out of the box in a clean Linux 16.04 with python 3.5. The code returns “TypeError: the JSON object must be str, not 'bytes'”. Based on a brief inspection of the source code, it seems that the json output of the urllib request is not being properly decoded. The authors might want to give that matter some thought, as a substantial amount of users might run into this issue. It could be as simple as explicitly defining the encoding. This type of early error tends to turn users off prematurely.

Secondly, the azure shiny app is also not working properly. When loading ‘user-provided images’, the server disconnects (on Chrome Version 79.0.3945.88 and Firefox 57.0.1). While the demo app should not preclude the publication of the manuscript, the authors might want to fix that prior to publication. At the very least, the demo app should be removed from the manuscript if not functional.

With regard to the main text, I thought the manuscript struck a good balance between attracting the interest of a more general audience and appeasing the more curious deep learning specialists. Having said that, it would be great if the authors could expand on the topic of data augmentation. I noticed that the source code does allow for data augmentation, but the manuscript was surprisingly mute about the topic. Since most users will rely on the fine-tuning of small datasets, data augmentation becomes critically important.

In general, this is a well written and user-friendly python package. It is also a remarkable example of open science. My comments here just aim to provide the authors with some (hopefully) useful feedback.