parameters.md

Table of contents

• Pipeline parameters
  • --images
  • --model
  • --masks
  • --label_spec
  • --multiscale
  • --chunksize
  • --ignore_senescence
  • --ignore_label
  • --outdir
  • --save_overlay
  • --save_mask
  • --save_rosette
  • --save_hull
  • --summary_diagnostics
  • --shiny

--images

Path to the images to be analysed. Supported image formats include PNG and JPEG.

Note that the path has to be enclosed in quotes and include a glob pattern that matches the images e.g. --images '/path/to/images/*png'

--model

Note that this parameter is ignored if --masks is set.

The pretrained model that is used for image segmentation to produce segmentation masks that are subsequently used to extract morphometric and colorimetric traits.

Pretrained models for semantic segmentation of leaf classes based on DeepLabv3+

Currently, there are 3 available models that will classify pixels based on the leaf classes they were trained on.

• --model 'A': trained on ground truth annotations for ordinary leaves (class_norm) only
• --model 'B': trained on ground truth annotations for ordinary (class_norm) and senescent (class_senesc) leaves
• --model 'C': trained on ground truth annotations for ordinary (class_norm), and senescent (class_senesc) and anthocyanin-rich (class_antho) leaves

Custom models

The pipeline can also use models that were trained with Deep Plant Phenomics to obtain segmentation masks that can then be used for trait extraction further downstream in the pipeline.

• --model 'DPP': uses the pretrained checkpoint of the Vegetation Segmentation model provided by DPP

Note that custom training checkpoints can be specified with the --dpp_checkpoint parameter. Currently, only 2-class models (background and plant) are supported.

--ignore_label

If using custom models or supplying already segmented masks, it can be desirable to ignore certain segmentation classes. Similar to --ignore_senescence, setting this parameter to the pixel value of the class to be ignored, will exclude those pixels from the calculation of morphometric traits.

--ignore_senescence

Note that this parameter only affects models B & C and will be ignored otherwise.

Ignore senescent class when calculating morphometric traits, focussing on living tissue only.

--multiscale

Note that this parameter is ignored if --masks is set.

Specifies whether the input image is scaled during model prediction. This yields higher accuracy at the cost of higher computational demand.

--masks

Path to grayscale segmentation masks corresponding to the images supplied with --images. When this parameter is set, the pipeline will not perform semantic segmentation using a trained model, it will only extract morphometric traits based on the user-supplied masks.

Note that the path has to be enclosed in quotes and include a glob pattern that matches the images e.g. --masks '/path/to/masks/*png'.

--label_spec

Note that this parameter is required if --masks is set and ignored otherwise.

Specifies a comma-separated list of key=value pairs of segmentation classes and their corresponding pixel values. Key is an arbitrary name and value is the corresponding grayscale pixel value in the supplied segmentation masks.

Note that the value of the background class has to be zero and the list has to be enclosed in quotes, e.g --label_spec 'class_background=0,class_norm=255'

--chunksize

The number of images in each chunk, determining the degree of parallelization. The smaller the chunksize, the more jobs will be spawned.

--outdir

The directory that results will be saved to.

--save_overlay

Save overlays of the original images with the segmentation masks to the results directory.

--save_mask

Save the segmentation masks to the results directory.

--save_rosette

Save rosette images that were cropped to the region of interest to the results directory.

--save_hull

Save convex hull images to the results directory.

--summary_diagnostics

Merge individual overlays, masks and rosette images into larger summaries that allow for quick inspection of results.

--shiny

Launch a Shiny app as the last step of the pipeline, allowing for interactive inspection of results.

Note that the app will run on the host where the main Nextflow process is running. If you are running the pipeline on a remote server, it has to expose port 44333 to the network.

If this is not the case, the shiny app can also be run manually on a local computer from the results directory. This has to happen from within the container or conda environment, respectively.

# change to results directory (needs to be copied to local file system first)
cd /path/to/results

# if using the conda environment
conda create -f /path/to/cloned/repository/containers/shiny/environment.yml
conda activate aradeepopsis-shiny
R -e "shiny::runApp('app.R', port=44333)"

# if using the container image
{docker|podman} run -v $(pwd):/mnt/shiny -p 44333:44333 quay.io/beckerlab/aradeepopsis-shiny:2.0 R -e "shiny::runApp('/mnt/shiny/app.R', port=44333, host='0.0.0.0')"

The shiny app can then be opened in a browser by typing localhost:44333 in the address bar. It will terminate when the browser window is closed.