- The generation of disease region and leaf instance masks in the existing dataset.
- An end-to-end deep learning framework with multitask loss function for simultaneous segmentation of leaf and diseased region using unified feature maps.
- To the best of our knowledge, this is the first study to quantify disease severity corresponding to individual leaves from UAV field images.
Data is labelled under the supervision of the experts. Kindly do cite if you use our data or find our results/code useful in your research:
@InProceedings{Garg_2021_WACV,
author = {Garg, Kanish and Bhugra, Swati and Lall, Brejesh},
title = {Automatic Quantification of Plant Disease From Field Image Data Using Deep Learning},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
month = {January},
year = {2021},
pages = {1965-1972}
}
An End to End Pipeline for Cascaded Instance Segmentation built on the top of MaskRCNN code.
Figure 1. Cascaded MRCNN framework. Given an input image (i) Primary mask branch predicts individual leaf segments (Leaf instance segmentation) and a (ii) Cascaded mask branch is added that utilises the feature maps corresponding to leaf instances for generating diseased region mask
Related Work: A Hierarchical Framework for Leaf Instance Segmentation: Application to Plant Phenotyping