An image analysis pipeline to measure quality-related traits of individual cranberries photographed in a homemade lightbox. These traits are being used for breeding and horticultural applications. This pipeline uses a fully convolutional network (FCN) workflow to perform semantic segmentation of berry pixels from background pixels. The FCN workflow was developed by Devin Rippner and the lightbox was constructed based on a similar application in hops.
fcn_model_training
-
resources and code to train developmental or production versions of the
FCN model.
resources
- other resources for the image analysis pipeline,
including a blank image for color correction.
lightbox
- materials and instructions for assembling your own
lightbox.
exampleImage
- three example images are located here.
imagesToSegment
- when deploying the FCN model in a production
environment, place images here for which inferences will be made.
output
- when deploying the FCN model, inference outputs will be
stored here.
Two scripts are available in this repository for deploying the FCN model under a production environment. Instructions for running either are contained within the notebook.
deploy_BerryBox_Production_FCNSegmentationModel_Colab.ipynb
- A
jupyter notebook for running the pipeline in a Google Colab environment.
deploy_BerryBox_Production_FCNSegmentationModel_SCINet.ipynb
- A
jupyter notebook for running the pipeline using SCINet HPC resources.
To deploy the trained, cranberry-specific production FCN model on Google Colab, follow these steps:
- Clone this repository using
git clone git@github.com:neyhartj/BerryBox.git
or download a .zip file of the repository. - Upload the repository to Google Drive.
- Download the trained model and metadata from this Google Drive folder. (This model will be made available elsewhere.) Upload both files to the uploaded BerryBox repository on Google Drive.
- Open the deploy_BerryBox_FCNSegmentationModel_Colab.ipynb notebook in Colab.
- Edit settings and variables as needed in the notebook. Do not edit
any lines after the “Other Settings” section. If you are using the
example images provided in this repository, the only setting you
should need to change is the
proj_dir
path. - Run the notebook!