The project is based on this paper: Adversarial Learning for Semi-Supervised Semantic Segmentation: https://arxiv.org/pdf/1802.07934.pdf
Source Code: https://github.com/hfslyc/AdvSemiSeg/tree/master
Data: https://www.kaggle.com/datasets/quadeer15sh/augmented-forest-segmentation
Pre-trained model's checkpoints are loaded from DrSleep's google drive. Please download the files and put them under model/pretrained.
Please put downloaded data into this directory: dataset/Forest
python3 train.py
You should be able to see a pop-up window of the last image from the testing result, and a loss and metric plot loss_metric.png in the main folder.
The batch size can be set in preprocess.py line 52.
The learning rate of the two optimizers can be set in train.py line 20 and 21.
The weight of the adversarial loss can be adjusted in train.py line 43 for training and line 90 for testing.