This is the official PyTorch implementation of “SADC-GAN: A Saliency-aware and Decomposition-consistent Generative Adversarial Network for Multimodal Image Fusion”
- PyTorch 1.7.0
- h5py 2.7.0
- numpy 1.19.5
- pillow 8.3.1
- torchvision 0.8.0
The datasets for training can be download from training_dataset.
Run:
python train_fusion_model.py --dataset_file=./train_datasets/MSRS_vis_inf_64.h5 --checkpoint_path=./trained_models/MSRS --epochs=30 --batch_size=96Run:
python train_fusion_model.py --dataset_file=./train_datasets/TNO_vis_inf_64.h5 --checkpoint_path=./trained_models/TNO --epochs=30 --batch_size=96Run:
python train_fusion_model.py --dataset_file=./train_datasets/Harvard_mri_pet_64.h5 --checkpoint_path=./trained_models/Harvard --epochs=30 --batch_size=96Run:
python test_fusion_model.py --dataset_path=./test_images/MSRS --hasRGB=Vis --save_path=./fusion_results/MSRS --checkpoint=./checkpoint/MSRS/fusion_model_G_MSRS.pthRun:
python test_fusion_model.py --dataset_path=./test_images/TNO --hasRGB=No --save_path=./fusion_results/TNO --checkpoint=./checkpoint/TNO/fusion_model_G_TNO.pthRun:
python test_fusion_model.py --dataset_path=./test_images/Harvard --hasRGB=Inf --save_path=./fusion_results/Harvard --checkpoint=./checkpoint/Harvard/fusion_model_G_Harvard.pthIf you have any questions, please create an issue.