# Neighbor2Neighbor: Self-Supervised Denoising from Single Noisy Images
**Tao Huang**, **Songjiang Li**, **Xu Jia**, **Huchuan Lu**, **Jianzhuang Liu**
**Abstract**:
_In the last few years, image denoising has benefited a lot from the fast development of neural networks. However, the requirement of large amounts of noisy-clean image pairs for supervision limits the wide use of these models. Although there have been a few attempts in training an image denoising model with only single noisy images, existing self-supervised denoising approaches suffer from inefficient network training, loss of useful information, or dependence on noise modeling. In this paper, we present a very simple yet effective method named Neighbor2Neighbor to train an effective image denoising model with only noisy images. Firstly, a random neighbor sub-sampler is proposed for the generation of training image pairs. In detail, input and target used to train a network are images sub-sampled from the same noisy image, satisfying the requirement that paired pixels of paired images are neighbors and have very similar appearance with each other. Secondly, a denoising network is trained on sub-sampled training pairs generated in the first stage, with a proposed regularizer as additional loss for better performance. The proposed Neighbor2Neighbor framework is able to enjoy the progress of state-of-the-art supervised denoising networks in network architecture design. Moreover, it avoids heavy dependence on the assumption of the noise distribution. We explain our approach from a theoretical perspective and further validate it through extensive experiments, including synthetic experiments with different noise distributions in sRGB space and real-world experiments on a denoising benchmark dataset in raw-RGB space._
**Official Pytorch implementation for the paper accepted by CVPR 2021.**
![Denoising comparison](img/readme_figure1.png "Denoising comparison")
![Denoising comparison](img/readme_figure2.png "Denoising comparison")
## Resources
- [Arxiv](https://arxiv.org/pdf/2101.02824.pdf)
- [Conference](https://openaccess.thecvf.com/content/CVPR2021/papers/Huang_Neighbor2Neighbor_Self-Supervised_Denoising_From_Single_Noisy_Images_CVPR_2021_paper.pdf)
- [Supplementary](https://openaccess.thecvf.com/content/CVPR2021/supplemental/Huang_Neighbor2Neighbor_Self-Supervised_Denoising_CVPR_2021_supplemental.pdf)
## Python Requirements
This code was tested on:
- Python 3.7
- Pytorch 1.3
## Preparing Training Dataset
Images in the training set are from the ImageNet validation set with size between 256x256 and 512x512 pixels.
There are 44328 images in total.
```bash
python dataset_tool.py
--input_dir=./ILSVRC2012_img_val
--save_dir=./Imagenet_val
```
- optional arguments:
- `input_dir` Path to the ImageNet validation set
- `save_dir` Path to save the training set
## Training
To train a network, run:
```bash
python train.py
--data_dir=./Imagenet_val
--val_dirs=./validation
--noisetype=gauss25
--save_model_path=./results
--log_name=unet_gauss25_b4e100r02
--increase_ratio=2
```
- selected optional arguments:
- `data_dir` Path to the training set
- `val_dirs` Path to the validation sets
- `noisetype` Distribution of image noise, choosing from `gauss25`, `gauss5_50`, `poisson30`, or `poisson5_50`
- `save_model_path` Base-path to the saved files
- `log_name` Path to the saved files
- `increase_ratio` Weight for the trade off between the reconstruction term and the regularization term in the loss function
## Citations
```
@InProceedings{Huang_2021_CVPR,
author = {Huang, Tao and Li, Songjiang and Jia, Xu and Lu, Huchuan and Liu, Jianzhuang},
title = {Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021},
pages = {14781-14790}
}
```
没有合适的资源?快使用搜索试试~ 我知道了~
neighbor2neighbor 代码 debug完,参考
共161个文件
png:136个
py:8个
pyc:8个
需积分: 0 30 下载量 18 浏览量
2022-11-22
09:20:28
上传
评论 1
收藏 62.87MB ZIP 举报
温馨提示
看有几个人问我neighbor2neighbor的代码问题,我直接把我debug好的代码发出来吧
资源推荐
资源详情
资源评论
收起资源包目录
neighbor2neighbor 代码 debug完,参考 (161个子文件)
LICENSE 2KB
training_script.md 6KB
README.md 4KB
kodim13.png 803KB
barbara.png 802KB
kodim08.png 770KB
kodim05.png 767KB
kodim18.png 763KB
kodim01.png 719KB
kodim24.png 690KB
kodim22.png 686KB
kodim14.png 676KB
kodim19.png 656KB
kodim04.png 622KB
kodim21.png 622KB
kodim11.png 606KB
kodim06.png 604KB
monarch.png 604KB
kodim02.png 604KB
kodim15.png 598KB
kodim17.png 588KB
kodim10.png 580KB
kodim09.png 569KB
kodim07.png 553KB
kodim23.png 545KB
kodim16.png 522KB
baboon.png 519KB
kodim12.png 519KB
pepper.png 496KB
kodim03.png 491KB
kodim20.png 481KB
zebra.png 433KB
175032.png 380KB
291000.png 376KB
lenna.png 366KB
33039.png 364KB
86016.png 359KB
55073.png 352KB
167083.png 344KB
flowers.png 340KB
41069.png 337KB
108070.png 334KB
229036.png 331KB
101085.png 331KB
156065.png 329KB
148026.png 328KB
175043.png 323KB
376043.png 322KB
69020.png 321KB
236037.png 321KB
65033.png 321KB
86068.png 320KB
304034.png 319KB
304074.png 318KB
148089.png 314KB
58060.png 313KB
38082.png 311KB
285079.png 309KB
134035.png 308KB
12084.png 303KB
69040.png 300KB
241048.png 299KB
85048.png 296KB
97033.png 291KB
130026.png 290KB
108005.png 288KB
16077.png 288KB
19021.png 286KB
89072.png 286KB
351093.png 285KB
223061.png 285KB
385039.png 284KB
253027.png 284KB
87046.png 284KB
208001.png 283KB
38092.png 283KB
159008.png 281KB
216081.png 281KB
105025.png 281KB
145086.png 278KB
271035.png 271KB
66053.png 271KB
76053.png 271KB
306005.png 269KB
157055.png 267KB
197017.png 264KB
ppt3.png 263KB
147091.png 261KB
86000.png 261KB
54082.png 261KB
170057.png 258KB
108082.png 257KB
196073.png 257KB
295087.png 254KB
21077.png 254KB
119082.png 253KB
69015.png 250KB
123074.png 250KB
163085.png 249KB
62096.png 248KB
共 161 条
- 1
- 2
资源评论
ssf-yasuo
- 粉丝: 166
- 资源: 4
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- 基于Python皮肤电信号的情绪识别算法源码+模型+PPT+详细文档+全部数据资料.zip
- 基于Python皮肤电信号的情绪识别算法源码+模型+PPT+详细文档+全部数据资料.zip
- 区块链毕业设计 基于区块链的慈善募捐平台的智能合约源码+详细文档+全部资料(高分项目).zip
- 区块链毕业设计 基于区块链的慈善募捐平台的智能合约源码+详细文档+全部资料(高分项目).zip
- 区块链毕业设计 基于区块链的慈善募捐平台的智能合约源码+详细文档+全部资料(高分项目).zip
- 免费听歌软件,支持无损音乐下载,支持收藏音乐同步
- 锐起RDV-5.0 build 6503
- ngrok-stable-windows-amd64.zip
- 狗脸检测数据集VOC+YOLO格式6154张1类别.7z
- 使用C++制作一个头像框
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功