Implementation of "High-Frequency Feature Transfer for Multispectral Image Super-Resolution" in PyTorch

• We conduct experiments on two remote sensing datasets: GaoFen-1 and GaoFen-2
• GaoFen-1: Contains 4,761 image pairs for training and 680 samples for testing.
• GaoFen-2: Contains 5,322 image pairs for training and 680 samples for testing. You can download the preprocessed datasets from here，then extract them to ' datasets/ '

Train teacher model using

python main.py -- Resume False --mode teacher --epochs epochs_number

The trained model is saved in ' checpoints/teacher/'

After obtaining the teacher models, please place the address of the saved teacher model on line 103 of main.py

if opt.Resume:
            path_checkpoint = './checkpoints/teacher/trained teacher model'

Then, train the student model using

python main.py --mode student --epochs epochs_number

The trained model is saved in ' checpoints/student/'

Put the address of the test model on line 134 of main.py

if opt.test:
        print("test process")
        #
        model_= torch.load('./checkpoints/student or teacher model/')

Evaluate the performance of the training model using

python main.py --mode student or teacher --test

We also provide our pretrained models on all four datasets for reference. You can download them from here.