We provide the pre-trained models, and the training code/data to reproduce the learning methods. This covers our experiments on synthetic super-resolution and real-world super-resolution.
Due to size issue and for anonymity, the pre-trained models will be uploaded to Google Drive later
To reproduce the regular training of:
- RCAN/RRDB/ESRGAN (net RCAN/RRDB/ESRGAN respectively):
python train.py --net RCAN
And to train with SFM (50% rate, central mode):
python train.py --net_mode RCAN --SFM 1
- KMSR
To after generate kernels in folder KMSR_kernel, run:
python train.py --net KMSR
Add --SFM 1 to train with SFM for the central mode of our SFM.
- IKC
We use the original IKC protocal. Similar to our other training codes, you can pass the SFM argument to the train_IKC.py file.
Due to size issue and for anonymity, the pre-trained models will be uploaded to Google Drive later
The same as the training process for synthetic dataset except for replacing the training data.