Một số thuật toán diffusion model for super resolution (LDM, ISSR và ESGAN-SwinIR)

List of algorithms for super resolution problem

1. LDM algorithm
   • It is one of the models of diffusion model.
   • Folder code: sources/LDM/taming-transformers/
   • Run the main function in main.py file (read carefully the parameters in README.md file)
   • Libraries that need to be installed should note: o pip install omegaconf o pip install einops o pip install pytorch_lightning==1.6.5 o pip install ipywidgets omegaconf>=2.0.0 pytorch-lightning>=1.0.8 torch-fidelity einops
   • Code implemented in ipynb (LDM.ipynb) for reference on how to run the superresolution model.
   • Link download: Link
2. ISSR algoritm (Image-Super-Resolution-via-Iterative-Refinement algorithm)
   • It is one of the models of diffusion model
   • Folder code: sources/ISSR/Image-Super-Resolution-via-Iterative-Refinement
   • Additional libraries that need to be installed: o pip install lmdb o pip install tensorboardx
   • Pretrained model in use: I830000_E32_opt
   • Run file sr.py to train. Run file infer.py to run the results.
   • Read more carefully in the README.md file in the root directory.
   • Code implemented using ipynb (ISSR.ipynb) to see how to infer data.
   • Link download: Link
3. SwinIR and ESRGAN algorithm
   • It is one of the GAN algorithms
   • Folder code: sources/SwinIR/Real-ESRGAN (root_path)
   • For SwinIR algorithm: root_path/SwinIR/ (swinir_root_path)
   • For ESRGAN, the root folder is root_path (esrgan_root_path)
   • Need to install some additional libraries: o !pip install basicsr o !pip install facexlib o !pip install gfpgan o !pip install -r requirements.txt o !pip install timm
   • For SwinIR, run file main_test_swinir.py (in folder swinir_root_path). Read more carefully the parameters at swinir_root_path/README.md
   • For ESRGAN model, run file inference_realesrgan.py of folder esrgan_root_path or file inference_realesrgan_video.py (if you want to test on video). You should read the file esrgan_root_path/README.md to better understand the parameters.
   • Pretrained models are saved in esrgan_root_path/pretrained_models
   • Code implemented with ipynb (esrgan_root_path /SwinIR_and_ESRGAN.ipynb) to see how to infer data.
   • Link download: Link

Link original source:

• https://github.com/JingyunLiang/SwinIR
• https://github.com/xinntao/ESRGAN
• Link
• https://github.com/IceClear/LDM-SRtuning/tree/main

Hết.