Spatial | 易学教程

ThunderNet ：像闪电一样，旷视再出超轻量级检测器，高达267fps | ICCV 2019

阅读更多关于 ThunderNet ：像闪电一样，旷视再出超轻量级检测器，高达267fps | ICCV 2019

> 论文提出了实时的超轻量级two-stage detector ThunderNet，靠着精心设计的主干网络以及提高特征表达能力的CEM和SAM模块，使用很少的计算量就能超越目前的one-stage detectors，在ARM平台也达到了实时性，GPU的速度更是达到267fps 来源：【晓飞的算法工程笔记】公众号论文: ThunderNet: Towards Real-time Generic Object Detection 论文地址： https://arxiv.org/abs/1903.11752 Introduction one-stage detector和two-stage detector分别有着实时优势和准确率优势。two-stage detector实时性较差，几乎不能在移动设备上运行，尽管已经有了light-head版本，但对于小主干网络而言，这依然是过度设计的。而one-stage由于缺少ROI-wise的特征提取，准确率一直较差为此，论文提出TunderNet，一个超轻量级的two-stage detector。在综合考虑输入分辨率，主干网络和detection head后，整体架构如图2，主要有两部分创新： Backbone part，提出轻量级主干网络SNet Detection part，参考Light-Head R-CNN的设计

【论文笔记 Detection】（2017 ICCV）Deformable Convolutional Network

阅读更多关于【论文笔记 Detection】（2017 ICCV）Deformable Convolutional Network

3 月，跳不动了？>>> Abstract I’ve writen many paper reviews in Chinese before, so to get more fun, English will be used for latter paper reviews gradually. A few days ago, we have reviewed STN . We all know that becaus of Pooling layer, CNN has spatial invariance(such as translation invariance, rotation invariance). And as the larger the pooling kernel and conv kernel are, invariance of CNN will get more and more powerful. But at the same time, the larger the pooling kernel and conv kernel are, CNN wil lost more and more local information. As a result, downsampling ratio needs to be adjusted

How to keep the only intersection of the spatial features & remove everything outside of a boundary?

阅读更多关于 How to keep the only intersection of the spatial features & remove everything outside of a boundary?

问题 I am trying to get rid of the spatial geometry that falls outside of the shapefile boundary I read. Is it possible to do this without manual software like Photoshop? Or me manually removing the tracts which span outside of the city's boundries. For example, I took out 14 tracts, this is there result: I have provided all of the subset of the data and the key to test it yourself. Code script is below, and the dataset is https://github.com/THsTestingGround/SO_geoSpatial_crop_Quest. I have done

PyTorch spatial transformer tutorial 空间变换网络(STN)教程

阅读更多关于 PyTorch spatial transformer tutorial 空间变换网络(STN)教程

要查看格式更完美的配图文章，请前去 http://studyai.com/pytorch-1.4/intermediate/spatial_transformer_tutorial.html 在本教程中，您将学习如何使用一种称为空间变换器网络的视觉注意机制来增强您的网络。您可以在 DeepMind paper 中更多地阅读有关空间变换器网络的内容。空间变换器网络(Spatial transformer networks, STN)是对任何空间变换的可微关注(differentiable attention)的推广。 STN允许神经网络学习如何对输入图像进行空间变换，以提高模型的几何不变性。例如，它可以裁剪感兴趣的区域、缩放和纠正图像的方向。这是一种有用的机制，因为CNN对图像旋转、尺度和更一般的仿射变换不具有不变性。关于STN最好的事情之一是能够简单地将它插入到任何现有的CNN中，而很少做任何修改。 from __future__ import print_function import torch import torch.nn as nn import torch.nn.functional as F import torch.optim as optim import torchvision from torchvision import datasets,

MCMCglmm ordinal model in R

阅读更多关于 MCMCglmm ordinal model in R

问题 I'm trying to explain changes in tree vitality from 1 to 3 (1=green, 2=damage, 3=dry) using climatic variables in an MCMCglmm model. Unfortunately, I am struggling with two questions: 1. How do I interpret the summary of the MCMCglmm model? I see which variables are significant, but what does it mean? Does this mean that with increasing precipitation the vitality gets to 1 (green)? I plotted the posterior means of the model and looks like precipitation and diameter have no influence on the

MCMCglmm ordinal model in R

阅读更多关于 MCMCglmm ordinal model in R

How to extract coordinates of colored dots from a jpeg image?

阅读更多关于 How to extract coordinates of colored dots from a jpeg image?

问题 I am trying to measure distances between objects of interests (in this example year rings in trees) using R. My earlier attempt was so complicated that I have difficulties reproducing the solution for a similar type of a problem using different kinds of figures. I think that there must be an easier way of doing the measurements. As nice as ImageJ might be for picture analysis, I find it too clumsy to use for repetitive work. Why not just to mark the objects of interest with different colors

How to filter an R simple features collection using sf methods like st_intersects()?

阅读更多关于 How to filter an R simple features collection using sf methods like st_intersects()?

问题 SF is the R-Spatial package designed to work with tidy syntax like dyplr and pipes. I would like to do a simple spatial filter on a simple features collection object. Given a simple features collection, I would like to return all features from the collection which meet some geometric condition. In particular, I would like to find the features that intersect with another object. SF provides the function st_intersects(x,y,...) to do this, but I cannot get it to work with dplyr. I'm using R 3.5

Adding a Polygon to a SpatialPolygonsDataFrame

阅读更多关于 Adding a Polygon to a SpatialPolygonsDataFrame

问题 How to add an polygon to a SpatialPolygonsDataFrame? Example: The script underneath will create a SpatialPolygonsDataFrame. I would like to add a polygon which consists of a large square around the existing polygons. library(rgdal) dsn <- system.file("vectors", package = "rgdal")[1] Scotland <- readOGR(dsn=dsn , layer="scot_BNG") plot(Scotland) Preferred end result: It is important that the rectangle becomes part of the SpatialPolygonsDataFrame. Since I have to do some calculation of the

Union all geometry in a SQL Server table like GeomUnion in Postgres

阅读更多关于 Union all geometry in a SQL Server table like GeomUnion in Postgres

问题 Just to clarify up-front: I'm talking about unioning geometry, not the SQL keyword UNION . I'm trying to move some spatial data from Postgres with PostGIS to SQL Server 2008. It was fine until I saw a statement like this: SELECT GeomUnion(the_geom) FROM some_table This unions all geometry in that column and return it as one result (similar to how COUNT works). As far I know, SQL Server only has the STUnion function, which unions one geometry with another. Is there any way to do something