Home

Awesome

DDSR

PyTorch code for paper "Dual-Diffusion: Dual Conditional Denoising Diffusion Probabilistic Models for Blind Super-Resolution Reconstruction in RSIs", which can be seen at https://doi.org/10.48550/arXiv.2305.12170. The code is based on https://github.com/megvii-research/DCLS-SR/tree/master/codes

The order of running code:

  1. RRDB_LR encoder (The pretrained RRDB_LR encoder has been given in the folder "Pretrained rrdb_LR encoder")
  2. Kernel Predictor
  3. HR reconstructor

Dataset

The link to the dataset: https://pan.baidu.com/s/1eD_mbFoNdPWfY8TCkjjfeA?pwd=j3vq

To transform datasets to binary files for efficient IO, run:

python codes/scripts/create_lmdb.py

To generate LRblur/LR_up/Bicubic datasets paths, run:

python codes/scripts/generate_mod_blur_LR_bic.py (You need to modify the file paths by yourself.)

Supplementary experiments

  1. How good is the proposed HR reconstructor compared to non-blind methods when all those methods are given the same (predicted or true) kernel？ image c26cbadd8186f509a09082d998bfd35 4476c2cebe3eb361a36cd86b2145242

  2. How good is the kernel predictor compared to the other blind methods' predictors 7ce23fae28bb9622b2d7f772b8cdcbc

    f8f284bdeaa819e46975db3a026e8a1