/rsimg-segmentation-pytorch

Pytorch-based semantic segmentation model for easy use on the remote sensing image

Primary LanguageJupyter NotebookGNU General Public License v3.0GPL-3.0

rsimg-segmentation-pytorch

We perform the more easily use of the semantic segmentation models on the remote sensing image with pytorch.
The user can perform the model training and validation by easily change the parameters in the scripts/config.py file.

Dataset

The surface water dataset built based on Sentinel-2 can be used for the quick testing in this repository, and the surface water dataset can be accessed at: https://zenodo.org/record/5205674.

Models to be achieved

  1. Unet (Simple)
  2. DeeplabV3Plus [Paper]
  3. DeeplabV3Plus with MobileNetV2 backbone
  4. WatNet [Paper]
  5. HRNet [Paper]
  6. GMNet [Paper]

Features

  1. Model training and validation Synchronously (scripts/trainer.py).
  2. Generate validation part from the whole dataset (notebooks/dset_val_patch.ipynb).
  3. Plot the metric figures (notebooks/metric_plot).