
Medical image registration
文章平均质量分 79
医学图像配准学习
海绵baby强无敌
CV菜鸡炼丹师
展开
-
Deep Learning Based Registration文章阅读(十六)《NccFlow: Unsupervised Learning of Optical Flow With Non-oc》
Deep Learning Based Registration文章阅读(十六)本次论文是2021.7月的文章,挂在arXiv,笔者还不知道发在了哪里,投的应该是IEEE Trans期刊,题目《NccFlow: Unsupervised Learning of Optical Flow With Non-occlusion from Geometry》。这篇文章的贡献是在non-occlusion的区域引入了几何约束,从而提升光流估计的准确度。在之前的无监督光流估计的工作中,引入了亮度一致性损失,但是亮度原创 2022-01-15 16:06:51 · 2237 阅读 · 0 评论 -
tensorflow实现Local Context Normalization
tensorflow实现Local Context Normalization参考代码:PyTorch implementation for Local Context Normalization: Revisiting Local Normalization参考文章:Local Context Normalization: Revisiting Local Normalization代码实现的是torch的code,以及是对2D图像的LCN,笔者改写成了tensorflow 1.4的code以及3D原创 2021-12-15 18:48:32 · 1801 阅读 · 1 评论 -
Deep Learning Based Registration文章阅读(十五)Learning by Distillation: A Self-Supervised Learning Framewo
Deep Learning Based Registration文章阅读(十五)《Learning by Distillation: A Self-Supervised Learning Framework for Optical Flow Estimation》论文来自TPAMI-2021年6月。论文提出了一种新的自监督蒸馏学习的估计光流的框架,以解决两个问题:1、目前大多数的(这篇文章写作时)全监督光流网络是使用合成数据来做预训练,并且如果要达到sota,一般要严格遵守不同合成数据集做预训练的训练数据原创 2021-11-25 15:25:41 · 1960 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(十四)Deformable MR-CT Image Registration Using an Unsupervised, D
Deep Learning Based Registration文章阅读(十四)本篇文章《Deformable MR-CT Image Registration Using an Unsupervised, Dual-Channel Network for Neurosurgical Guidance》来自于MIA 2021,本篇文章处理多模态问题还是基于GAN来做,通过结合CycleGAN网络来将多模态问题转为单模态问题来解。整体框架还是无监督,NCC做转为单模态后的similarity loss,个人原创 2021-11-17 14:52:00 · 950 阅读 · 0 评论 -
3D Slicer auto W/L实现
3D Slicer auto W/L实现参考博客:3Dslicer1:入门及基本控制自动窗宽窗位的一些思路How auto W/L is implemented in 3DSlicer?python代码:根据How auto W/L is implemented in 3DSlicer?实现// auto W/Ldef auto_wl(img, low=0.1, high=0.99): imhist, bins = np.histogram(img.flatten(), int(ma原创 2021-09-26 17:19:48 · 342 阅读 · 0 评论 -
python图像序列转为git动图
python图像序列转为git动图python代码:// make gifimport imageioimport osfrom functools import cmp_to_keydef compare(num1, num2): if int(num1) > int(num2): return 1 elif int(num1) == int(num2): return 0 else: return -1ima原创 2021-09-26 17:11:36 · 302 阅读 · 0 评论 -
多模态医学图像数据集
多模态医学图像数据集参考博客:数据集:一文道尽医学图像数据集与竞赛医学影像系列:一 数据集合集 最新最全医学图像数据集汇总医学影像数据集集锦医学影像开源数据集脑的数据一般可以多模态,MRI-T1/T2 etc, CT/MRI,单独的配准图像应该很难找,一般是找分割的多模态图像。上海交通MedMNIST10 datasets, single modal, classification taskLink:https://arxiv.org/pdf/2010.14925.pdfMedPix原创 2021-07-24 00:39:44 · 14035 阅读 · 46 评论 -
Deep Learning Based Registration文章阅读(十三)UPFlow: Upsampling Pyramid for Unsupervised Optical Flow Lea
Deep Learning Based Registration文章阅读(十三)本次文章是CVPR2021 megvii关于无监督光流的一篇,孙剑通讯。Motivation目前的无监督光流的sota是UFlow,整合了目前为止包括pyramid structure等各个模块后形成的框架。但是目前的pyramid structure有两个问题,这篇文章也是根据这两个问题提出了相应的method解决从而取得无监督sota。第一,pyramid structure中存在upsampling的操作,但是目前原创 2021-07-04 18:05:20 · 1041 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(十二)《AutoFlow: Learning a Better Training Set for Optical Flow》
Deep Learning Based Registration文章阅读(十二)这次的文章是CVPR2021关于光流的一篇文章《AutoFlow: Learning a Better Training Set for Optical Flow》。光流的学习可以分为有监督和无监督两类,对于有监督的学习,通常需要flow field的ground truth。真实世界中通常很难获得这样的ground truth,所以一个目前常用的训练方法是用合成数据。现在最常用的合成数据是Flying Chairs和Flyi原创 2021-07-04 16:46:38 · 698 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(十一)《A Coarse-to-fine Deformable Transformation Framework For U》
Deep Learning Based Registration文章阅读(十一)本次文章是TMI的文章《A Coarse-to-fine Deformable Transformation Framework For Unsupervised Multi-contrast MRImage Registration With Dual Consistency Constraint》。总体感觉方法的创新性不是很高,但是做的很规矩很完整。Motivation本文的motivation主要是说MR图像有很多原创 2021-06-18 17:51:40 · 782 阅读 · 2 评论 -
Deep Learning Based Registration文章阅读(十)《Learning Optical Flow from a Few Matches》
Deep Learning Based Registration文章阅读(十)本次文章是CVPR2021年《Learning Optical Flow from a Few Matches》。基于深度学习的光流法的研究一直是CV领域坚持不懈的追求。目前取得sota的RAFT(Recurrent all-pairs field transforms for optical flow)需要计算dense correlation volume,这会带来很大的memory和computation的负担,并且对于f原创 2021-05-14 22:27:46 · 763 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(九)《ViT-V-Net: Vision Transformer for Unsupervised Volumetric M》
Deep Learning Based Registration文章阅读(九)本次文章是一篇arXiv上的短文《ViT-V-Net: Vision Transformer for Unsupervised Volumetric Medical Image Registration》,首次把Transformer用于3D医学图像配准中。Transformer作为NLP领域目前大放异彩的结构,自2020年10月被首次用到cv分类任务中后,现在在cv各个任务中,包括细粒度分类,目标检测,分割等都取得了比CNN更原创 2021-04-30 16:44:55 · 2053 阅读 · 2 评论 -
Deep Learning Based Registration文章阅读(八)《Learning optical flow from still images》
Deep Learning Based Registration文章阅读(八)本次文章来自于CVPR2021的《Learning optical flow from still images》,关于光流的文章还有几篇,为《Learning Optical Flow from a Few Matches》、《Upsampling Pyramid for Unsupervised Optical Flow Learning》等。本篇文章的题目比较吸引人,从静态图像学光流,笔者以为是把SIFT flow的思想原创 2021-04-15 23:07:57 · 711 阅读 · 1 评论 -
Deep Learning Based Registration文章阅读(七)《CycleMorph: Cycle consistent unsupervised deformable image r
Deep Learning Based Registration文章阅读(七)本次文献是MIA 2021年3.21号最新的文章《CycleMorph: Cycle consistent unsupervised deformable image registration》。本篇文章的网络名字沿用了VoxelMorph的格式,但是不是VoxelMorph团队做的。这篇文章的baseline沿用了VoxelMorph并加以改进,主要是引入了一个cycle consistent,这个约束类似于cycleGAN。原创 2021-04-08 17:43:08 · 1381 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(六)《Adversarial learning for mono- or multi-modal registration》
Deep Learning Based Registration文章阅读(三)《Adversarial learning for mono- or multi-modal registration》是MIA 2019的一篇文章,是做原创 2021-03-31 16:56:59 · 565 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(五)《Anatomy-guided Multimodal Registration by Learning Segment 》
Deep Learning Based Registration文章阅读(五)这篇文章是MIA2021新出的一篇文章《Anatomy-guided Multimodal Registration by Learning Segmentation without GroundTruth: Application to Intraprocedural CBCT/MR Liver Segmentation and Registration》。目的是做多模态3D的affine registration。关于为什么原创 2021-03-31 16:20:48 · 915 阅读 · 1 评论 -
Deep Learning Based Registration文章阅读(四)《End-to-end multimodal image registration via reinforcement 》
Deep Learning Based Registration文章阅读(三)这次的文章是MIA上的一篇文章《End-to-end multimodal image registration via reinforcement learning》,与之前DL based registration不同,这篇文章是通过强化学习来解决配准问题,不过目前只能解决仿射变换。因为强化学习是通过当前的评价函数来决策下一步的action,对于deformable的配准自由度很大,下一步的决策空间比较大,而仿射变化的自由度原创 2021-03-19 15:45:44 · 632 阅读 · 0 评论 -
Deep Learning Based Registration文章阅读(三)《Reinforced Feature Points: Optimizing Feature Detection and》
Deep Learning Based Registration文章阅读(三)《Reinforced Feature Points: Optimizing Feature Detection and Description for a High-Level Task》是CVPR2020年的文章。先上总体直观效果:RootSIFT是一个SIFT的归一化的变种,使得效果更加鲁棒,SuperPoint是目前在low-level task上state-of-art performance的框架。最下面是这篇文章提原创 2021-03-19 13:04:08 · 625 阅读 · 1 评论 -
Deep Learning Based Registration文章阅读(二)《Unsupervised Multi-Modal Image Registration via Geometry Pre
Deep Learning Based Registration文章阅读(二)本次阅读的文章题为《Unsupervised Multi-Modal Image Registration via Geometry Preserving Image-to-Image Translation》,来源CVPR2020。Motivation多模态图像配准一直是配准领域一个比较棘手的问题,主要是因为不同模态图像的intensity distribution不同,从而使得单模态配准的基于intensity的simi原创 2021-03-06 14:29:57 · 1922 阅读 · 4 评论 -
Deep Learning Based Registration文章阅读(一)《Content-Aware Unsupervised Deep Homography Estimation》
笔者每周计划详细看2-3篇比较新的深度学习做配准的文章,主要来源是CV比较好的期刊及顶会CV常用期刊及网址,看完后记录一下印象及理解会更加深刻,更欢迎交流~Deep Learning Based Registration文章阅读(一)本次阅读的文章题为《Content-Aware Unsupervised DeepHomography Estimation》,来源ECCV2020。Prepare Knowledge读者对单应性估计以及文中提到的传统方法中用到的RANSAC都不熟悉,故根据文中参考文原创 2021-02-23 17:02:04 · 1996 阅读 · 0 评论 -
二维图像配准flow field可视化
二维图像配准flow field可视化笔者最近要对二维图像的flow field做可视化,参考的文章是FlowNet1.0中supplemetary中的方案,该方案和SIFT Flow是一样的,SIFT Flow提供了matlab代码,下载链接 Matlab Flow可视化(审核还未通过),笔者改写成了python代码。可视化策略FlowNet1.0文章 文章链接 描述了Flow Field可视化策略Python代码// visualize flow fielddef visualize_f原创 2021-01-10 13:18:06 · 1248 阅读 · 1 评论 -
ICCV2019《Recursive Cascaded Networks for Unsupervised Medical Image Registration》代码学习
ICCV2019《Recursive Cascaded Networks for Unsupervised Medical Image Registration》代码学习该论文代码为tensorflow,本学习过程最终目的是将该tensorflow代码改写为pytorchGithub代码链接:Recursive Cascaded Networks for Unsupervised Medical Image Registrationeval.py跑通1、按照要求将所需val datasets以及pr原创 2020-12-02 23:11:23 · 3169 阅读 · 7 评论