深度学习论文笔记(七)---Deconvolution network Learning Deconvolution Network for Semantic Segmentation Author: ②目标的细节结构常常丢失或者被平滑处理掉, 所以输入deconvolution-layer的label map就太粗糙了,而且deconvolution 这个步骤在FCN这篇文章中做的过于简单了。 缺少一个在大量数据上得到训练的deconvolution network使得准确地重构物体边界的高维非线性结构变得困难。 针对上面的两个limitations,这篇文章提出的贡献有: • We learn a deep deconvolution network, which is composed of deconvolution 所以就要进行下一步操作,deconvolution ②Deconvolution ? 如图所示,Deconvolution的细节我就不描述了。简而言之,它的功能就和convolution相反。
Deconvolution大致可以分为以下几个方面: (1)unsupervised learning,其实就是covolutional sparse coding[1][2]:这里的deconv只是观念上和传统的 网络结构还是一样有deconvolution ?
https://blog.csdn.net/zhangjunhit/article/details/72528610 Learning Deconvolution Network for procedure 太粗糙太简单,FCN 的 deconvolution procedure输入尺寸只有16 × 16,将这个尺寸通过 bilinear interpolation 放大到输入图像尺寸 corresponds to feature extractor 反卷积网络根据特征产生分割结果 deconvolution network is a shape generator that Deconvolution Network for Segmentation 反卷积网络中主要有两个操作步骤: unpooling and deconvolution 3.2.1 Unpooling unpooling layers 得到一个放大的但是稀疏的响应特征图, 这里通过deconvolution layers 来将稀疏的特征变为稠密的特征 The deconvolution
Learning a perspective-embedded deconvolution network for crowd counting 没有找到代码 本文在人群密度估计这个问题上的创新点: fuse the perspective into a deconvolution network 首先看看 Perspective Perspective is an inherent property Deconvolution network CFCN-DCN:加了两个卷积层 conv5 with filter size 5 × 5 and conv6 with filter size 7 × Perspective fusion the perspective-embedded deconvolution network (PE-CFCN-DCN) 这里看 图2 比较直接明了 A Each fusion layer is inserted before each deconvolution block for guided interpolation. ? ? ?
目录 写在前面 什么是deconvolution convolution过程 transposed convolution过程 transposed convolution的计算 整除的情况 不整除的情况 开篇先上图,图为deconvolution在像素级语义分割中的一种应用,直观感觉deconvolution是一个upsampling的过程,像是convolution的对称过程。 本文将深入deconvolution的细节,并通过如下方式展开: 先回答 什么是deconvolution? 什么是deconvolution 首先要明确的是,deconvolution并不是个好名字,因为它存在歧义: deconvolution最初被定义为“inverse of convolution”或者“ 本文谈论的是deconvolution的第2个含义,后面统一使用transposed convolution这个名字。 什么是transposed convolution?
转置卷积也被称作: “分数步长卷积(Fractionally-strided convolution)“和”反卷积(Deconvolution)”. Evan Shelhamer, Trevor Darrell https://people.eecs.berkeley.edu/~jonlong/long_shelhamer_fcn.pdf [4] Deconvolution
/deconvolution_res_transcrip").mkdir(exist_ok=True)deconvolution.add_deconvolution_results_to_dataset (stdata=stdata_filtered, result=deconvolution_result)deconvolution_result.save('. /deconvolution_res_transcrip/deconvolve_result.h5')deconvolution.plot_cell_num_scatterpie( stdata /deconvolution_plots_transcrip/Scatter_piechart.png')deconvolution.plot_cell_prob( stdata=stdata_filtered /deconvolution_plots_transcrip', output_format= 'png',)deconvolution.plot_cell_num( stdata
Libraries of Post-Xenium Human Colon Cancer (FFPE)'# Define the number of topics to import from the deconvolution data for the specified number of topicsthis_deconv_name = deconv_folders_name + '/deconvolution_k' + No deconvolution data has been added to the domain.')# Create a mask for spots that are in the tissuespot_mask datacluster_id = []deconv = {}# Setting up the deconvolution dictionary with empty lists for each topicfor = 'Barcode': deconv[col_name] = []# For each barcode ID, find the associated cluster and deconvolution
然而,基于测序的空间转录组学尚未达到单细胞分辨率,因此需要先进的解卷积(deconvolution)方法来推断每个空间位置上的细胞类型贡献。 文章链接:Gaspard-Boulinc, L.C., Gortana, L., Walter, T. et al.Cell-type deconvolution methods for spatial Methods are grouped by the frameworks used to perform deconvolution. EnDecon stands alone in its category as the only consensus method among the other deconvolution methods Summary of deconvolution method characteristics across six categories: use of reference, use of extra
of the PASCAL classes (including background) at each of the coarse output locations, followed by a deconvolution layer to bilinearly upsample the coarse outputs to pixel-dense outputs 得到一个 10 × 10 的分割结果A,我们使用一个 deconvolution layer 进行双线性上采样到输入图像尺寸得到 FCN-32s分割结果, 直接放大32倍 deconvolution layer 中的滤波器参数通过学习得到。 A stack of deconvolution layers and activation functions can even learn a nonlinear upsampling. 然后 再对 FCN-16s中 的分割结果 C 进行 2× upsampling layer 得到一个放大2倍的分割结果图C2, 将这两个分类置信度图求和相加得到了 一个分割结果图 E,最后使用一个 deconvolution
).as_list() deconv_shape = [conv_shape[0], conv_shape[1]*2, conv_shape[2]*2, conv_shape[3]] deconvolution , output_shape=deconv_shape, strides=[1, 2, 2, 1], padding='VALID') return in_put, convolution, deconvolution input_x = tf.placeholder(dtype=tf.float32, shape=[1, 4, 4, 1]) in_put, convolution, deconvolution = sess.run([in_put, convolution, deconvolution], feed_dict={input_x:np.random.uniform(low=0, high=255 # print _deconvolution if __name__ == "__main__": main() 2017-09-29 09:51:41.472842
联合单细胞进行空间转录组spot注释的方法,本文介绍下20202年发表于NBT的文献Spatially informed cell type deconvolution for spatial transcriptomics spatial_loca <- Brain_ST@images$slice1@coordinates spatial_location <- spatial_loca[,2:3] #名字必须是x y ,不然后面CARD_deconvolution 2.2,CARD 解卷积 使用CARD_deconvolution函数解卷积 CARD_obj = CARD_deconvolution(CARD_object = CARD_obj) ## create CARD_obj = CARD_deconvolution(CARD_object = CARD_obj) ## Select Informative Genes! . ## Deconvolution Starts! ... ## Deconvolution Finish!
提到的算法主要是2大类,包括: 基于GSEA的半定量方法 Deconvolution algorithms(分为partial deconvolution和complete deconvolution) partial deconvolution Deconvolution algorithms可以理解为一个基因在样本中的表达量是该基因在样本中不同细胞亚群表达水平和细胞分数权重的线性组合。 complete deconvolution 相比较与partial deconvolution,complete deconvolution不仅可以估计相对细胞分数同时还能disentangle表达谱 Deconvolution algorithms其中一个挑战就是对这些unknown tumor content的稳健性。EPIC和quanTIseq允许最终细胞分数总和低于1,来估计未知细胞分数。 此外,多重共线性也是deconvolution algorithm要考虑的问题,如基因表达在related cells中的高度相关。
Intelligence Charting Tissue Expression Anatomy by Spatial Transcriptome Decomposition Advances in mixed cell deconvolution enable quantification of cell types in spatially-resolved gene expression data Colour deconvolution: Deep learning–based cell composition analysis from tissue expression profiles Updating immune cell deconvolution spotlight_deconvolution函数接受scRNAseq数据和空间转录组学数据计数矩阵的Seurat对象,并返回反卷积结果。 ? spotlight_deconvolution spotlight_ls <- spotlight_deconvolution(se_sc = cortex_sc,
interpretable framework for solving a wide range of image reconstruction problems such as denoising and deconvolution state-of-the-art methods on all these tasks and is significantly faster than optimization-based approaches to deconvolution space of predicted filters for different tasks. keywords: inverse problem, spatially-variant blind deconvolution
AlexNet, VGG, GoogLeNet) fine-tune, pixel-to-pixel input of any size, output classification maps(heatmap) deconvolution 之后对得到的1*1*1000的输出,做upsampling(deconvolution)得到和原图一样大小的输出,所有输出合并之后得到如上图所示的heat map 当然这里作者的deconvolution 会相对变小,可能会损失全局信息,且会对卷积层引入更多运算 对于第一种方法,虽然receptive fileds没有变小,但是由于原图被划分成f*f的区域输入网络,使得filters无法感受更精细的信息 deconvolution
spatialType = "QualityControl", spatialFeatures=c("UMI","Gene") ) 图片 Deconvolve ST data 通过两个阶段,SpaCET.deconvolution 图片 # deconvolve ST data SpaCET_obj <- SpaCET.deconvolution(SpaCET_obj, cancerType="BRCA", coreNo=8) # show the ST deconvolution results SpaCET_obj@results$deconvolution$propMat[1:13,1:6] ## # further deconvolve malignant cell states SpaCET_obj <- SpaCET.deconvolution.malignant(SpaCET_obj, malignantCutoff = 0.7, coreNo = 8) # show cancer cell state fraction of the first five spots SpaCET_obj@results$deconvolution
interpretable framework for solving a wide range of image reconstruction problems such as denoising and deconvolution state-of-the-art methods on all these tasks and is significantly faster than optimization-based approaches to deconvolution space of predicted filters for different tasks. keywords: inverse problem, spatially-variant blind deconvolution
这里介绍了三种特征图融合的方式: 1) pooling, 从前往后增加特征图,归一化尺寸 2) 借鉴图像分割中的 deconvolution, 从后往前增加特征图,归一化尺寸 3)本文的 Rainbow concatenation=pooling+deconvolution,这样每一个尺度的特征图数量都是一样的,这就导致后面可以使用一个分类器在不同尺度上检测 By using
反卷积(Deconvolution)与棋盘效应(Checkerboard Artifacts) Deconvolution and Checkerboard Artifacts