Code and models for "Rethinking Deep Image Prior for Denoising" (ICCV 2021)

Last update: Dec 29, 2022

Related tags

Overview

DIP-denosing

This is a code repo for Rethinking Deep Image Prior for Denoising (ICCV 2021).

Addressing the relationship between Deep image prior and effective degrees of freedom, DIP-SURE with STE(stochestic temporal ensemble) shows reasonable result on single image denoising.

If you use any of this code, please cite the following publication:

@article{jo2021dipdenoising,
  author  = {Yeonsik Jo, Se young chun,  and Choi, Jonghyun},
  title     = {Rethinking Deep Image Prior for Denoising},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month     = {October},
  year      = {2021},
  pages     = {5087-5096}
}

Working environment

TITAN Xp
ubuntu 18.04.4
pytorch 1.6

Note: Experimental results were not checked in other environments.

Set-up

Make your own environment

conda create --name DIP --file requirements.txt
conda avtivate DIP
pip install tqdm

Inference

Produce CSet9 result

bash exp_denoising.sh CSet9 <GPU ID>

For your own data with sigma=25 setup

mkdir testset/<YOUR_DATASET>
python main.py --dip_type eSURE_new --net_type s2s --exp_tag <EXP_NAME> --optim RAdam --force_steplr --desc sigma25   denoising --sigma 25 --eval_data <YOUR_DATASET>

Browsing experimental result

We provide reporting code with invoke.

invoke showtable csv/<exp_type>/<exp_tag>

Example.

invoke showtable csv/poisson/MNIST/
PURE_dc_scale001_new                     optimal stopping : 384.30,     31.97/0.02      | ZCSC : 447.60,         31.26/0.02 | STE 31.99/0.02
PURE_dc_scale01_new                      optimal stopping : 94.70,      24.96/0.12      | ZCSC : 144.60,         24.04/0.14 | STE 24.89/0.12
PURE_dc_scale02_new                      optimal stopping : 70.30,      22.92/0.20      | ZCSC : 110.00,         21.82/0.22 | STE 22.83/0.20
<EXEPRIMENTAL NAME>                      optimal stopping :<STEP>,      <PSNR>/<LPIPS>  | ZCSC : <STEP>,      <PSNR>/<LPIPS>| STE <PSNR>/<LPIPS>

The reported numbers are PSNR/LPIPS.

Results in paper

For the result used on paper, please refer this link.

SSIM score

For SSIM score of color images, I used matlab code same as the author of S2S.
This is the demo code I received from the S2S author.
Thank you Mingqin!

% examples
ref = im2double(imread('gt.png'));
noisy = im2double(imread('noisy.png'));
psnr_result = psnr(ref, noisy);
ssim_result = ssim(ref, noisy);

License

MIT license.

Contacts

For questions, please send an email to [email protected]

Code and models for "Rethinking Deep Image Prior for Denoising" (ICCV 2021)

Related tags

Overview

DIP-denosing

Working environment

Set-up

Inference

Browsing experimental result

Results in paper

SSIM score

License

Contacts

Owner

Computer Vision Lab. @ GIST

Revisiting Weakly Supervised Pre-Training of Visual Perception Models

automatic color-grading

Lighting the Darkness in the Deep Learning Era: A Survey, An Online Platform, A New Dataset

Realtime YOLO Monster Detection With Non Maximum Supression

A torch.Tensor-like DataFrame library supporting multiple execution runtimes and Arrow as a common memory format

Uni-Fold: Training your own deep protein-folding models

Medical-Image-Triage-and-Classification-System-Based-on-COVID-19-CT-and-X-ray-Scan-Dataset

This is an official repository of CLGo: Learning to Predict 3D Lane Shape and Camera Pose from a Single Image via Geometry Constraints

Implementation of Auto-Conditioned Recurrent Networks for Extended Complex Human Motion Synthesis

PAthological QUpath Obsession - QuPath and Python conversations

AWS documentation corpus for zero-shot open-book question answering.

Code base for "On-the-Fly Test-time Adaptation for Medical Image Segmentation"

Implementation of Convolutional LSTM in PyTorch.

Moer Grounded Image Captioning by Distilling Image-Text Matching Model

Paddle pit - Rethinking Spatial Dimensions of Vision Transformers

Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models

Baseline of DCASE 2020 task 4

object recognition with machine learning on Respberry pi

NVTabular is a feature engineering and preprocessing library for tabular data designed to quickly and easily manipulate terabyte scale datasets used to train deep learning based recommender systems.

⚓ Eurybia monitor model drift over time and securize model deployment with data validation