# High-Resolution Leaf Image Sequences with Geometric Alignment for Dynamic Phenotyping of Foliar Diseases

## TL;DR

This paper introduces a dataset of high-resolution wheat leaf images with disease symptoms, enabling detailed study of plant disease progression and better image analysis methods.

## Contribution

The novel contribution is a large, geometrically aligned leaf image dataset with tools for studying dynamic foliar disease development.

## Key findings

- A dataset of 12,520 high-resolution wheat leaf images with disease symptoms is presented.
- Geometric alignment of images achieves a median precision of 0.16 mm.
- The dataset includes segmentation masks and metadata for studying disease dynamics.

## Abstract

Time-resolved phenotyping of disease symptoms enables dissection of resistance mechanisms and improves diagnosis, but acquiring phenotypic data at satisfactory scale remains challenging. Advances in imaging and image processing have improved measurement precision, robustness, and throughput, but further improvements are needed for practical application. We present a data set comprising 12,520 high-resolution (~0.03 mm/pixel) RGB images representing 1,032 time series of wheat leaves with developing disease symptoms. All images are geometrically aligned with a median precision of 0.16 mm (≈5 pixels). The dataset includes transformation matrices, symptom segmentation masks, metadata on treatments, weather, crop phenology, and disease occurrence, and a lightweight Python toolkit for loading, aligning, inspecting, and editing image sequences. These resources enable detailed investigation of leaf-level disease dynamics such as lesion, pustule, and fruiting body emergence rates, lesion growth, and dynamic interactions of disease development with spatial and environmental contexts. They offer a broad basis for developing improved methods for image alignment and symptom detection, segmentation, and tracking, possibly by tackling these connected challenges within a single end-to-end framework.