James Ball

Suggestions to improve clarity of tutorial

1

- [x] check for missing function calls - `project_to_geojson`, `stitch_crowns`, `clean_crowns` - [ ] clarify how to connect with personal google drive - [x] clarify `Confidence_scores` and how filtering should...

Annoying small polygons seem to be produce as an artefact of converting the mask to a polygon. Some area based filtering has already been applied but this is not ideal...

Additional training on varied landscapes will produce more transferable model

It needs to only create an output directory when training is being done - not necessary for landscape level predictions

Invalid polygon geometries can cause errors in evaluation

Fix test loader - why does it select just the first test

1

Does testing look at all test data when this has [0]?? https://github.com/PatBall1/detectree2/blob/08851c8d5e30cd178836a30311ee95ccb0bb7b44/detectree2/models/train.py#L149

Upgrade system so that multiple classes can be predicted rather than the simple single class delineation ('tree')

Introduce data readers that allow multispectral images to be used in training and prediction. There will be a possible issue with using pre-trained base models as has been done previously....

This seems a little out of date: https://patball1.github.io/detectree2/installation.html Recently got the package working on our cluster with this sequence: ``` conda create -n dtree2 python==3.10 conda activate dtree2 conda install...

Function needs updating for multispectral landscape predictions