xuru_0927的博客

原创 Constrained Probabilistic Mask Learning for Task-specific Undersampled MRI Reconstruction

【代码】Constrained Probabilistic Mask Learning for Task-specific Undersampled MRI Reconstruction。

原创 UserWarning: FigureCanvasAgg is non-interactive, and thus cannot be shown

【代码】UserWarning: FigureCanvasAgg is non-interactive, and thus cannot be shown。

原创 install GitHub Desktop on ubuntu

【代码】install GitHub Desktop on ubuntu。

原创 UNet笔记

import ostask_name = 'Task114_heart_MNMs'raw_data_base_path="/home/ilkay/Documents/ruru/nnUNet/ocmr_task114/nnUNet_raw"preprocessed_path="/home/ilkay/Documents/ruru/nnUNet/ocmr_task114/nnUNet_preprocessed"result_folder_path="/home/ilkay/Documents/ruru

原创 进展---一些笔记

把OCMR的fs的shape写进csv文件import numpy as npimport mathfrom ismrmrdtools import transformimport ocmr_unet.read_ocmr as readimport osfiles = sorted(list(pathlib.Path('/home/ilkay/Documents/ruru/pg_mri/datasets/ocmr/train/').iterdir()))out_list = []for f

原创 SeqMRI train process

SeqMRI train processLOUPE大体流程train的过程从初始化mask到得到zero-filled imagereconstruction过程怎么得到masksequential网络LOUPE大体流程LOUPE( (samplers): ModuleList( (0): LOUPESampler( (gen_mask): LineConstrainedProbMask() ) ) (reconstructor): LOUPEUNet(

原创 DL-HW

DL homeworkHW0linear_regression:HW0linear_regression:https://www.cs.toronto.edu/~frossard/post/linear_regression/

原创 c1_greedy_policy_processing

def get_policy_probs(model, recons, mask): #mask(3,1,1,256,1) recons(3,1,256,256) channel_size = mask.shape[1] #1 res = mask.size(-2) #256 recons = recons.view(mask.size(0) * channel_size, 1, res, res) #(3,1,256,256) # Obtain policy model

原创 可视化代码

可视化代码real转complexcomplex转realdpireal转complexdef real2complex(x): x = x[:,0] + (1j * x[:,1]) return xcomplex转realdef complex2real(x): ''' Parameter --------- x: ndarray assumes at least 2d. Last 2D axes are split in te

原创 pg_mri----OCMR policy network

before policy network作为initialstep0经过policy network输出变量output是(1,255)，然后排出中间的白条，取其余的最大值位置第145。因为8个trajectories，这里复制出了8个channel得到(8,256)，每个都是(255,255),所以8个相同的mask_kspacestep1这里actions = policy.sample()tensor([[144, 144, 146, 144, 144, 146, 144, 146]].

原创 pg_mri----brain policy network

train_policy.pyCreate data loadertest_loader = create_data_loader()test_data_range_dict = create_data_range_dict()do_and_log_evaluation一直往mask中加竖线，for循环16遍step=0step=1step=2,3......15循环完所有batch计算ssim和psnrCreate data loadertest_loader = create_data_loader

原创 pg_mri----brain reconstruction

train-reconstructiontrain_loader = create_data_loader(args, 'train', shuffle=True)dev_loader = create_data_loader(args, 'val')display_loader = create_data_loader(args, 'val', display=True)scheduler = torch.optim.lr_scheduler.StepLR(optimizer, args.lr_step_

原创 我遇到的疑惑及解惑

我遇到的疑惑及解惑Why Zero-padding值得反复看的视频fft, ifft， fftshift，ifftshiftWhy Zero-paddingZero-padding methods to improve the DFT值得反复看的视频MRI Image Parameters and OptimizationActive MR k-space Sampling with Reinforcement Learning https://arxiv.org/abs/2007.10469

原创 cascade MRI reconstruction:evaluate

if draw: pred = pred[0] grnd = grnd[0] err = np.abs(pred - grnd) ** 2 err -= err.min() err /= err.max() err *= 255 err = err.astype(np.uint8) err = cv2.applyColorMap(err, 2) plt.imsave(os.path.join(kwargs['output_path'],

原创 model.load_state_dict(state_dict)报错问题

看一下训练得到的模型参数 state_dict = torch.load('logs/sanity_3/checkpoint', map_location='cuda' if args['train']['cuda'] else 'cpu') state_dict = state_dict['model']看一下参数for k,v in state_dict.items(): print(k)输出：module.block.0.layers.0.weightmodul.

原创 MSE & PSNR & SSIM

PSNR & SSIMcascade MRI reconstruction: PSNRbase_psnrtest_psnrsummarycascade MRI reconstruction: SSIMUnderstanding Regression - Mean Squared Error / LossIn Part 1 of this video, calculation of MSE is explained and in the following Part 2 of the vide

原创 cascade MRI reconstruction:loss function

MSELOSS大体流程手动计算粗略手动计算调torchstep by step手动计算用numpysummary总结数据集值扩大1000倍大体流程用了10个.mat，让train为6个，test为4个，batch_size=2loss = criterion(output['image'], batch['full'])执行loss跳到这里再跳到这里再跳到这里再跳到这里跳到这里执行上面的return语句时跳到这里然后return结束返回到原位置跳回到原来的

原创 iTerm2

iTerm2Oh My ZshThemeszsh Auto SuggestionsSyntax HighlightingWord JumpsChange the username@hostname终端运行.ipynbexport https_proxy=http://127.0.0.1:7890 http_proxy=http://127.0.0.1:7890 all_proxy=socks5://127.0.0.1:789Oh My Zshhttps://ohmyz.shInstall with

原创 2021-07-26科研进展

The words in blue are the knowledge points involved in the presentationIn preprocessing stage：First, the original data is preprocessed using zero-paddingThen the fully sampled original data is accelerated to obtain the undersampled dataIn order to per.

原创 Parallel MRI

Contents4.1 Introduction4.2 Parallel Imaging as an Inverse Problem .4.2.1Forward Model ………4.2.2Image Reconstruction ………4.2.3Regularization……4.2.4Discretization………………4.2.5Numerical Optimization …4.3 Autocalibration ……4.3.1 Nonlinear Inverse Reconstruction .

原创 Compressed Sensing in Quantitative MRI

Contents3.1 Introduction3.2 Principles of Compressed Sensing3.3 Generalized Formalism for an MR Parameter Mapping Model-Based: Reconstruction3.3.1 Sampling Pattern Consideration for Model-Based Reconstruction3.3.2 Model-Based Sparsity Constraint3.3.3 Apply

原创 Data Driven Methods for MRI Reconstruction

Code: JuliaIntroductionHistoryscopeMeasurement modelbrief review of classic methodsOrdinary least-squaresSmooth regularizationSparsity regularizers：Basicsparsity models：synthesis formProximal methodssparsity regularizers：AdvancedNon- SENSE methodsGRAPPA /

原创 cascade MRI reconstruction: train.py--3

output = model(batch)class CascadeNetwork(nn.Module): def forward(self, x): return self.block(x)batch=x#第一个ResnetBlockclass ResnetBlock(nn.Module): def forward(self, x): residual = x['image'] out = self.layers(x['image

原创 pytorch - connection between loss.backward() and optimizer.step()

Some answers explained well, but I’d like to give a specific example to explain the mechanism.Suppose we have a function : z = 3 x^2 + y^3.The updating gradient formula of z w.r.t x and y is:initial values are x=1 and y=2.x = torch.tensor([1.0], req

原创 k-space相关

k-space相关Zero-fillingZero-fillingProcessing of MRI data is usually performed in an array processor using the Fast Fourier Transform (FFT) algorithm. These computational requirements generally require the input data to be in a matrix whose dimensions are

原创 cascade MRI reconstruction: cartesian_mask--2

def cartesian_mask(self, shape): N, Nx, Ny = int(np.prod(shape[:-2])), shape[-2], shape[-1] pdf_x = normal_pdf(Nx, 0.5/(Nx/10.)**2) # shape：256pdf_x.reshape(16,16)lmda = Nx/(2.*self.acc) # 32n_lines = int(Nx / self.acc) # 64# ad...

原创 cascade MRI reconstruction: dataset.py--1

首先配置文件不能少config.yaml# Model Parametersnetwork: num_cascades: 6 num_layers: 5 # Number of layers in the CNN per cascade num_filters: 64 kernel_size: 3 stride: 1 padding: 1 #A padding of 1 is needed to keep t

原创 HW1_ML---Data Wrangling with pandas

每个代表几个sample，一列代表一个feature有三行sample: 1, 2, 3。有三个feature：a, b, c

原创 2021-3-22科研进展

对于FastMRI数据集Kneepath = r'/Volumes/exterD/codeTemp/pg_mri/datasets_knee/singlecoil_train/file1000026.h5'target = h5py.File(path, 'r')['reconstruction_esc']Output：<HDF5 dataset "reconstruction_esc": shape (30, 320, 320), type "<f4">The invers

原创 2021-3-8科研进展

原创 2021-2-22论文Experimental design for MRI by greedy policy search解读

**下面这是我的2021-2-22 presentation**ContentsPart 0----Basic theoryPart 1-----NetworkPart 2-----Policy gradients and reinforcement learningPart 3-----MRI subsamplingPart 4-----MethodPart 0----Basic theory下面的图：The central part of a filtering m...

原创 MRI数据集分析

MRI数据集分析Fast MRI数据集Knee MRI数据集分析5个items一一show出来Brain MRI数据集分析三个itemsAWS OCMR数据集分析Fast MRI数据集if you would like to download the files, you will need to submit a request on https://fastmri.med.nyu.edu:Knee MRI数据集分析path = r'/Volumes/exterD/codeTemp/pg_mri

原创 IEEE paper写作总结LaTeX格式

最近写了篇IEEE paper，被LaTeX折磨的死去活来总算完工，下面把我用到的LaTeX语法做个笔记，以后肯定也会用到https://github.com/xr0927/chapter10-IEEE-paper-LaTeX-format感觉有用的话顺便给个star⭐⭐...

原创 比赛总结------津南数字制造算法挑战赛——赛场二：物流货物限制品监测

比赛总结------津南数字制造算法挑战赛——赛场二：物流货物限制品监测求star⭐⭐????????https://github.com/xr0927/project5–Monitoring-of-restricted-goods-in-logistics...

原创 比赛总结-----------华为云人工智能大赛·垃圾分类挑战杯

比赛总结求star⭐⭐????????https://github.com/xr0927/project4–Garbage-sorting

原创 我会不定期翻译一些CVPR等等视觉方面的论文，不定期更新上传到这里

我会不定期翻译一些论文，不定期更新上传到这里https://github.com/xr0927/chapter8-translate-papers已翻译完：1.CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting-----------------...

原创 CNN-based Cascaded Multi-task Learning of High-level Prior and Density Estimation for Crowd Counting

crowdcount-cascaded一步步跟我学训练和测试https://github.com/xr0927/chapter7-Learning-Cascaded-pytorch顺便给美女个star哦star⭐⭐

原创 跟我一步步学CSRNet训练和测试

https://github.com/xr0927/chapter5-learning_CSRNet给美女个star哦⭐⭐

原创 CVPR2019-Context-Aware_Crowd_Counting论文解读

CVPR2019一篇关于基于上下文感知的人群计数，训练和测试都已经跑通，我就顺便写了一篇博客，以代码的形式对网络架构进行了剖析https://github.com/xr0927/chapter9-learnCVPR2019-Context-Aware_Crowd_Counting顺便给美女个star哦star⭐...

原创 CondaHTTPError HTTP 404 NOT FOUND for url

CondaHTTPError: HTTP 404 NOT FOUND for url <https://mirrors.ustc.edu.cn/anaconda/cloud/msys2/noarch/删除原来镜像源conda config --remove-key channels重新添加镜像源conda config --add channels https://mirrors....