Deep Learning based High-Throughput Phenotyping of Chalkiness in Rice Exposed to High Night Temperature

Dataset of original images is available here. Individual rice seed images are extracted from original images.
Datasets of individual polished and unpolished seed images are available here. Each image is annotated in terms of chalkiness. We provide train, development, test subsets.
Models for polished and unpolished rice are available here. The models are used in main.py file.
Web-based application:https://rootanatomy.cs.ksu.edu/html_rice/

The repository contains data and models for detecting chalkiness in rice grains. The models are based on a weakly supervised segementation approach, Grad-CAM (Selvaraju et al., 2017). The Pytorch implementation of Grad-CAM (WonKwang Lee, 2018), available at https://github.com/1Konny/gradcam_plus_plus-pytorch, was used to segment chaliness area from rice grains. We experiment with models that are trained with both polished and unpolished rice seeds. The dataset, model, and code are available in this repository to enable reproducibility, reuse and transfer of knowledge to chalkiness detection tasks for other types of rice seeds.

Content of the Repository

main.py is the code for segementing the rice chalkiness area from rice seeds

Segement rice chalkiness area from rice seeds:

See below steps to segement rice chalkiness area from rice seeds
Step1: change the path in line 15 and line 26
Step2: python main.py
Step3: segmentation result saved at ./outputs