Semantic Image Segmentation of Eucalyptus Trees in Panoramic Imagery: Approaches of Deep Learning for Forest Management
This work assesses the performance of novel deep-learning methods for eucalyptus tree segmentation in panoramic RGB ground-level images. We applied four models: the FCN, GCNet, ANN, and PointRend, using a challenging dataset composed of eucalyptus trees with variation in distances between trunks, curvature, sizes, and trunks of different diameters. The four semantic segmentation methods were trained and evaluated in five-fold cross-validation. A quantitative-qualitative analysis is presented, along with a discussion about the advantages and limitations of each CNN applied.
The master branch works with PyTorch 1.3+.
The datasets of the models used in this work are available at https://drive.google.com/drive/folders/19-xnaSfwRppgkPlfLazYeBMXI3cxY_sJ?usp=sharing and should be placed in the dataset folder.
The checkpoints of the models used in this work are available at https://drive.google.com/drive/folders/19-xnaSfwRppgkPlfLazYeBMXI3cxY_sJ?usp=sharing and should be placed in the checkpoints folder.
This project is released under the MIT license.
Please refer to get_started.md for installation and dataset_prepare.md for dataset preparation.
Please see train.md and inference.md for the basic usage of MMSegmentation. There are also tutorials for customizing dataset, designing data pipeline, customizing modules, and customizing runtime. We also provide many training tricks.
If you find this project useful in your research, please consider cite:
@unpublished{carvalho2024semantic,
author = {Mário de Araújo Carvalho and José Marcato Junior and Amaury Antônio de Castro Junior and Celso Soares Costa and Pedro Alberto Pereira Zamboni and José Augusto Correa Martins and Lucas Prado Osco and Michelle Taís Garcia Furuya and Felipe David Georges Gomes and Ana Paula Marques Ramos and Henrique Lopes Siqueira and Diogo Nunes Gonçalves and Jonathan Li and Wesley Nunes Gonçalves},
title = {Semantic Image Segmentation of Eucalyptus Trees in Panoramic Imagery: Approaches of Deep Learning for Forest Management},
note = {Submitted to Ecological Informatics, under review},
year = {2024},
url = {https://github.com/MarioCarvalhoBr/segmentation-eucalyptus-trees-in-panoramic-images}
}
We appreciate all contributions to improve MMSegmentation. Please refer to CONTRIBUTING.md for the contributing guideline.
This project is based on the project - MMSegmentation: OpenMMLab semantic segmentation toolbox and benchmark. MMSegmentation is an open source project that welcome any contribution and feedback. We wish that the toolbox and benchmark could serve the growing research community by providing a flexible as well as standardized toolkit to reimplement existing methods and develop their own new semantic segmentation methods. Documentation of MMSegmentation is available at https://mmsegmentation.readthedocs.io.