We found it during previous examples working with this data that the leaf
and leafscan
content types
led to the highest quality returns from the system. This repository is designed to create COCO-type masks
to train the Mask R-CNN (either TensorFlow or PyTorch). Leaf mask creation is designed to be the first step
of a plant identification process.
The primary dataset chosen for this is the LifeCLEF 2015 Plant Task https://www.imageclef.org/lifeclef/2015/plant which is primarily based on plant species in Europe. The link includes more detail on the task and the datset itself.
The primary motivation for this project is towards building an invasive species
detection
system which more naturally integrates with photo apps on a persons phone for after / during
hiking identification and notification of invasive plant species. These models are a first
step towards a more (read actual) system to perform that task.
I will include installation instructions at a later date, high level, download the
2015 training data and create a symbolic link to the data in this folder labeled
data
. A bit on the nose, but it works.