Trained Weights for Herbarium Sample Segmentation CNN Model

This dataset contains the trained weights for a convolutional neural network (CNN) model designed to segment herbarium samples from scanned images. The CNN model is built on the OCRNet architecture, as implemented in the https://github.com/PaddlePaddle/PaddleSeg toolkit for semantic segmentation.
The final weights were obtained by fine-tuning the OCRNet architecture on a dataset of 400 high-resolution fern images, originally published by White et al. (2020; https://doi.org/10.1002/aps3.11352). This dataset includes corresponding masks generated through a combination of automated and manual curation tools, providing a robust foundation for training a segmentation model.
For instructions on using the model, please visit: https://github.com/sales-lab/powerplant.git