- Clone repo
git clone https://github.com/MoooJianG/LDCSR.git- Install dependencies
conda create -n LDCSR python=3.10
conda activate LDCSR
pip install -r requirements.txtWe support AID, DOTA, and DIOR out‑of‑the‑box. Download HR images using the official links below (or your own data) and generate LR/HR pairs via bicubic down‑sampling.
| Data Type | AID | DOTA | DIOR |
|---|---|---|---|
| HR | HR | None | None |
| LR | LR | LR | LR |
# Example: split AID
python data/prepare_split.py --split_file AID_split.pkl --data_path dataset/RawAID --output_path dataset/AIDCustom datasets should replicate the following folder structure:
└── dataset
└── YourData
├── Train
| ├── HR
| └── LR
├── Test
└── Val
# First-stage
python train.py --config configs/first_stage_kl_v6.yaml
# Second-stage
python train.py --config configs/second_stage_van_v4.yamlpython test.py --checkpoint path/to/checkpoint.ckpt --datasets AID --scales 4If you have any questions or suggestions, feel free to contact me.
Email：20220119004@bfsu.edu.cn
@ARTICLE{11006698,
author={Wu, Hanlin and Mo, Jiangwei and Sun, Xiaohui and Ma, Jie},
journal={IEEE Transactions on Geoscience and Remote Sensing},
title={Latent Diffusion, Implicit Amplification: Efficient Continuous-Scale Super-Resolution for Remote Sensing Images},
year={2025},
volume={},
number={},
pages={1-1},
keywords={Diffusion models;Training;Image synthesis;Noise reduction;Visualization;Decoding;Computational modeling;Remote sensing;Image reconstruction;Autoencoders;Remote sensing;super-resolution;latent diffusion;continuous-scale},
doi={10.1109/TGRS.2025.3571290}}