Skip to content

Latest commit

 

History

History
68 lines (53 loc) · 2 KB

README.md

File metadata and controls

68 lines (53 loc) · 2 KB

MOB-GCN: Multiscale Object-Based Graph Neural Network for Hyperspectral Image Segmentation and Classification

MOB-GCN Architecture

Contributors:

  • Tuan-Anh Yang
  • Truong-Son Hy (PI)
  • Phuong Dao (PI)

Data

We use six benchmark HSI datasets to evaluate our approach, which have the following characteristics. INDIAN, SALINAS, PAVIA, BOTSWANA, KENNEDY can be found at Hyperspectral Remote Sensing Scenes and TORONT (UT-HSI-301) can be found at University of Toronto HSI-301 Dataset.

Code

Requirements

Create virtual environment

py -m venv .venv
.venv\Scripts\activate

Installing modules

pip install -r requirements.txt

Run

Load Experiments with Training (GCN)

py experiment_gcn.py --dataset INDIAN --segmentation_size 10 --training

Load Experiments without Training, assuming trained before (GCN)

py experiment_gcn.py --dataset INDIAN --segmentation_size 10

Inference (GCN)

py inference_gcn.py ---dataset INDIAN --segmentation_size 10 --weights_path output/INDIAN/experiment/gcn_model.pth --output_path output/INDIAN

Load Experiments with Training (MOB-GCN)

py experiment_mgn.py --dataset INDIAN --segmentation_size 10 --training

Load Experiments without Training, assuming trained before (MOB-GCN)

py experiment_mgn.py --dataset INDIAN --segmentation_size 10

Find optimal scales for MOB-GCN inference

py optimal_scale.py --dataset INDIAN --segmentation_size 10

Inference (MOB-GCN)

py inference_mgn.py ---dataset INDIAN --segmentation_size 10 --weights_path output/INDIAN/experiment/gcn_model.pth --output_path output/INDIAN --num_clusters 33,28,22,13,4

Benchmarking GCN, MOB-GCN (with num_classes) and MOB-GCN (with optimal scales)

py benchmark.py --dataset INDIAN --segmentation_size 10 --sample_size 0.05 --num_clusters 33,28,22,13,4

Citation