Flattened Leaves

This guide provides shows you how to evaluate images of flattened leaves outdoors following the method of Zenkl et al. 2025a and Anderegg et al. 2024, visualize results and compute disease metrics.

First make sure that all your images are exported to a common image format (.jpg, .png, etc.), they are named uniquely, and cropped to a consistent resolution. For images acquired according to Anderegg et al. 2024 we recommend 2048 x 8192 px.

Then you can use an existing configuration and predict simply a complete folder. You can either use a flat or nested folder structure.

Using a flat folder structure for export:

from leaf.inference import Predictor

# 'flattened_leaves' uses 2048 x 8192 px images
pred = Predictor(config_name='flattened_leaves')
pred.predict(images_src='data/images', export_dst='export')

This code snippet produces all necessary predictions in the export folder for evaluating images of canopy. If you want to make visualizations for manual inspection of the images execute the following:

from leaf.visualization import FlattenedVisualizer

vis = FlattenedVisualizer(
src_root='export',
rgb_root='data/images',
export_root='export',
)
vis.visualize()

Now you should see new subfolders in the export directory with rendered visualizations for each individual results and for the combined predictions.

Note

Visulization of predictions currently takes significantly longer than the actual inference. Consider visualizing only a subset of the data.

Now you should have the following structure:

export/
├── symptoms_det/
│   ├── pred/
│   └── vis/
├── symptoms_seg/
│   ├── pred/
│   └── vis/
└── visualization_symptoms/

Regardless of the visualization step you can compute the relevant disease metrics. To compute the relevant disease metrics per image execute the following snippet

from leaf.metrics import flat_leaves_evaluation_wrapper

flat_leaves_evaluation_wrapper(root_folder='export', results_path='export/flattened_results.csv')