计算机视觉论文阅读笔记集合(持续更新)

论文阅读笔记集合


深度学习环境配置

Anaconda介绍、安装及使用教程
Linux系统下conda的安装与使用
linux非root下安装CUDAhttps://developer.nvidia.com/cuda-80-ga2-download-archive
非root用户在Linux系统下安装cuda
Linux之Anaconda环境下安装TensorFlow
windows10+cuda+pytorch
如何解压tar.gz文件或是.gz文件

查看gpu使用情况: nvidia-smi
查看谁在用gpu: ps aux | grep PID
测试显卡是否能用:

import tensorflow as tf	
print(tf.test.is_gpu_available())

import torch
print(torch.cuda.is_available())

部分论文阅读笔记

1.图像修复综述

Image Restoraion: From Sparse and Low-rank Priors to Deep Priors

图像恢复笔记

2. 三权加边稀疏编码方案

A Trilateral Weighted Sparse Coding Scheme for Real-World Image Denoising
徐君主页
Github:https://github.com/csjunxu/TWSC-ECCV2018

交替方向乘子法(ADMM)算法
硬阈值(Hard Thresholding)函数解读
软阈值(Soft Thresholding)函数解读
西尔维斯特方程_百度百科
正定矩阵与半正定矩阵

3. DL图像增强方法

DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks

Github:https://github.com/aiff22/DPED
效果演示:http://www.vision.ee.ethz.ch/~ihnatova/

DL图像增强方法--《DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks,2017》
CNN 图像增强--DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks
【论文详解】DPED:DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks
《DSLR-Quality Photos on Mobile Devices with Deep Convolutional Networks》研读笔记

输入是手机照片(比如iphone, blackberry等),输出是单反照片。
重点之一:使用gan网络
重点之二:增加了新的loss function。

  • color loss:颜色损失计算之前,要对图像进行高斯模糊处理。为什么用高斯模糊处理的原因是,高斯模糊能出去高频的信息,使得颜色之间更容易比较。颜色损失对小的误差匹配有较高的容忍性。因此,能学习到和目标图片相似的颜色分布。
  • texture loss:使用gan网络,用判别网络的正确率来作为质地误差的衡量。
  • content loss:以vgg抽象出的高维特征的欧式距离作为内容损失。和上面采用gram matrix不同。
  • total variation loss:目的是获得较为平滑的输出。

4. 盲运动去模糊:

DeblurGAN: Blind Motion Deblurring Using Conditional Adversarial Networks
Github: https://github.com/KupynOrest/DeblurGAN

拍照手抖有救了!
论文笔记
论文翻译

5. 高级视觉任务:

When Image Denoising Meets High-Level Vision Tasks: A Deep Learning Approach
Github: https://github.com/Ding-Liu/DeepDenoising

文章学习37
知乎笔记

6. DnCNN降噪

Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising
Github: https://github.com/cszn/DnCNN

文献翻译
学习笔记
论文详解

7. 3D点云去噪

Denoising Point Sets via L0 Minimization
Robust Feature-Preserving Denoising of 3D Point Clouds

主成分分析(PCA)原理详解:降维算法
RANSAC算法详解:采用迭代的方式从一组包含离群的被观测数据中估算出数学模型的参数

论文详细笔记:

8. 时变点云的降噪

GRAPH-BASED DENOISING FOR TIME-VARYING POINT CLOUDS

数据集与打开软件:https://lts2.epfl.ch/research/reproducible-research/graph-based-point-cloud-denoising/

9. 图神经网络(调研/综述)

GRAPH-BASED DENOISING FOR TIME-VARYING POINT CLOUDS

Github: https://github.com/thunlp/GNNPapers (GNN论文列表)

【图神经网络】清华大学孙茂松组一文综述 GNN
图神经网络GNN-Graph Neural Networks: A Review of Methods and Applications
图神经网络综述:模型与应用

10. Vgg风格迁移

  1. Deep Photo Style Transfer (2017)
    GitHub 1: luanfujun/deep-photo-styletransfer(Torch)
    GitHub 2: LouieYang/deep-photo-styletransfer-tf(Tensorflow)

深度卷积神经网络图像风格变换
GitHub上大热的Deep Photo终于有TensorFlow版了
从零开始在 Google 云平台上免费跑深度图像风格转换工具——Deep Photo Style Transfer
Deep Photo Styletransfer的一种纯Tensorflow实现,教你如何转换图片风格

  1. Image Style Transfer Using Convolutional Neural Networks (2016)
    GitHub: MingtaoGuo/Style-transfer-with-neural-algorithm

卷积神经网络图像风格转移
使用卷积神经网络的图片风格转换器
Image Style Transfer:多风格 TensorFlow 实现
【啄米日常】5:keras示例程序解析(2):图像风格转移
vgg网络结构

11. WCT风格迁移

  1. Universal Style Transfer via Feature Transforms (2017)
    GitHub: Yijunmaverick/UniversalStyleTransfer

wct|大专栏
《Universal Style Transfer via Feature Transforms》论文解读
【两分钟论文】在这个AI眼里,万物皆可“迁移”

  1. A Closed-form Solution to Photorealistic Image Stylization (2018)
    GitHub: NVIDIA/FastPhotoStyle

PhotoWCT:A Closed-form Solution to Photorealistic Image
Stylization

  1. Photorealistic Style Transfer via Wavelet Transforms (2019)
    GitHub: clovaai/WCT2

风格迁移论文"Photorealistic Style Transfer via Wavelet Transforms"

12. 图像修复

PEPSI++: Fast and Lightweight Network for Image Inpainting (2020)
Github: Forty-lock/PEPSI-Fast_image_inpainting_with_parallel_decoding_network

论文阅读-PEPSI : Fast Image Inpainting with Parallel Decoding Network
[论文笔记] PEPSI : Fast Image Inpainting with Parallel Decoding Network
Hinge loss

13. Generative Inpainting

  1. Generative Image Inpainting with Contextual Attention (2018)

[AI] 论文笔记 - CVPR2018: Generative Image Inpainting with Contextual
Attention 【论文译文】Generative Image Inpainting with Contextual Attention

  1. Free-Form Image Inpainting with Gated Convolution (2019)
    Github: https://github.com/JiahuiYu/generative_inpainting

论文阅读及代码运行--Free-Form Image
Inpainting with Gated Convolution
论文 | Free-Form Image Inpainting with Gated Convolution
[译] Free-Form Image Inpainting with Gated Convolution
论文笔记《Free-form image inpainting with gated convolution》
【CNN基础】局部响应归一化、gated convolution和gated deconv

掩码处理: 【T-DT视觉组】3.19 放射变换透视变换教学

其他方法:
youyuge34/Anime-InPainting (效果差)
KumapowerLIU/CSA-inpainting (没跑)

14. 即插即用(去噪、去模糊、超分辨)

Deep Plug-and-Play Super-Resolution for Arbitrary Blur Kernels (2019)
Github: cszn/DPSR

适用于任意模糊内核的深度即插即用超分辨率(DPSR论文笔记-2019CVPR)

15. 各种GAN

Github: https://github.com/eriklindernoren/PyTorch-GAN

(哎。懒得更了)


posted @ 2021-09-04 19:45  埕同学  阅读(353)  评论(0编辑  收藏  举报