alfred_torres-CSDN博客

原创 cuda_knn vs torch.topk()

发现一个快速的KNN gpu算法KNN_CUDADGCNN KNN之前的knn算法用的都是DGCNN里面的torch.topk(),但是效率很低，网络训练既占用显存又慢；代替的pointnet2中的ball_query方法又不是严格的knn，可能会导致训练精度变低。### DGCNN knndef knn(x, k): inner = -2*torch.matmul(x.transpose(2, 1), x) xx = torch.sum(x**2, dim=1, keepdim=

2022-03-21 10:02:48 2185 4

原创 [点云处理] 两篇文章：PointAtrousNet和ECG

PointAtrousNet: Point Atrous Convolution for Point Cloud AnalysisECG: Edge-aware Point Cloud Completion with Graph Convolution两篇文章的作者都是Liang Pan，都发表在了IEEE ROBOTICS AND AUTOMATION LETTERS上，一个是2019年，一个是2020年，这个期刊可以考虑投。PointAtrousNet: Point Atrous Convolut

2020-09-23 15:47:51 908

原创 [点云识别]-What You See is What You Get: Exploiting Visibility for 3D Object Detection

What You See is What You Get: Exploiting Visibility for 3D Object Detection你看到的才是你能get：探究可见性对3D物体检测的影响CVPR 2020，对目前使用的3D数据结构进行了改造，增加了可见性这一属性，对于遮挡物体的检测和自动驾驶具有重大意义。摘要原文译文Recent advances in 3D sensing have created unique challenges for computer

2020-09-17 20:47:53 626

原创 [点云分割]-Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds

Multi-Path Region Mining For Weakly Supervised 3D Semantic Segmentation on Point Clouds一种用于3D点云弱监督语义分割的多路径区域搜索方法CVPR 2020本文的亮点有两个：1对点云语义分割用弱监督训练，2 Multi-Path Region Mining模块，主要挖掘了attention机制在点云中的应用摘要原文译文Point clouds provide intrinsic geometr

2020-07-30 22:07:37 452

原创 [点云补全]-Point Cloud Completion by Skip-attention Network with Hierarchical Folding

Point Cloud Completion by Skip-attention Network with Hierarchical FoldingCVPR 2020利用多级Folding结构和skip-attention来进行点云补齐Folding结构是之前就有的，本文主要是利用skip-attention将folding结构堆叠起来，更加深了，有点像受到了ResNet和DeepGCNs的启发。摘要原文译文Point cloud completion aims to infe

2020-07-22 16:29:57 3152

原创 [点云场景语义]-Anisotropic Convolutional Networks for 3D Semantic Scene Completion

Anisotropic Convolutional Networks for 3D Semantic Scene CompletionAnisotropic表示各向异性CVPR 2020摘要原文译文As a voxel-wise labeling task, semantic scene completion (SSC) tries to simultaneously infer the occupancy and semantic labels for a scene from

2020-07-21 19:58:02 527

原创 [点云目标检测]-A Hierarchical Graph Network for 3D Object Detection on Point Clouds

A Hierarchical Graph Network for 3D Object Detection on Point CloudsCVPR 2020摘要原文译文3D object detection on point clouds ﬁnds many applications.3D点云物体检测有越来越多的应用.However, most known point cloud object detection methods did not adequately acco

2020-07-20 17:04:40 1809

原创 [点云识别]-SegGCN: Efficient 3D Point Cloud Segmentation with Fuzzy Spherical Kernel

SegGCN: Efficient 3D Point Cloud Segmentation with Fuzzy Spherical Kernel摘要Fuzzy kernelSegGCN实验部分kernel compareRobustness to point density (missing data)Decoder module choiceCVPR 2020主要是由2019CVPROctree guided cnn with spherical kernels for 3d point cloud

2020-07-13 20:09:28 1667

原创点云识别-Geometry Sharing Network for 3D Point Cloud Classification and Segmentation

2020 AAAI一种关注几何特征的点云分类和Part分割网络摘要原文译文In spite of the recent progresses on classifying 3D point cloud with deep CNNs, large geometric transformations like rotation and translation remain c...

2020-01-31 20:43:07 1917

原创点云识别-Learning to Sample

Learning to Sample2019 CVPR摘要原文译文Processing large point clouds is a challenging task. Therefore, the data is often sampled to a size that can be processed more easily.处理大规模点云是一项具有挑战性的任务，...

2019-12-03 17:24:28 2477 1

原创 3D 点云识别: Geometric Feedback Network for Point Cloud Classification

Geometric Feedback Network for Point Cloud Classification用于点云分类的几何特征反馈网络2019/12/2 arXiv摘要As the basic task of point cloud learning, classification is fundamental but always challenging....

2019-12-02 17:03:21 955

原创 [人脸实战]-RK3399 基于ncnn实现vgg-face

基于ncnn的人脸识别系统目标：在RK3399板子上实现2维人脸识别平台：NanoPC-T4板子，系统基于Ubuntu Desktop 18.04的FriendlyDesktop准备工作：按照wiki上的教程装opencv3.41 安装ncnnncnn是腾讯出品的一个为手机端极致优化的高性能神经网络前向计算框架。ncnn git地址1.1 编译按照git上的教程【Build for ...

2019-04-09 16:26:24 3141

原创人脸识别-3D:Disentangled Representation Laerning for 3D Face Shape

Disentangled Representation Learning for 3D Face Shape探索3D人脸形状的表示学习2019 CVPR 中科大张举勇

2019-03-01 22:23:34 2349

原创论文速看[2019-1-17]-Support Vector Guided Softmax Loss for Face Recognition

一篇关于人脸识别中loss改进的方法文章i地址：https://128.84.21.199/abs/1812.11317作者github地址：https://github.com/xiaoboCASIA/SV-X-Softmax还未完成

2019-01-17 21:26:09 856

原创论文速看[2019-1-10] Low-Cost Transfer Learning of Face Tasks

一篇关于人脸识别迁移学习的文章IIT出品，文章地址：Arxiv摘要Do we know what the different filters of a face network represent? Can we use this filter information to train other tasks without transfer learning? For instance,...

2019-01-15 16:13:08 207 4

原创论文速看[2019-1-7]-Multi-scale tensor voting for feature extraction from unstructured point clouds

一篇关于从无结构化点云特征提取的文章2012 Graphical Models摘要Identifying sharp features in a 3D model is essential for shape analysis, matching and a wide range of geometry processing applications. This paper present...

2019-01-07 09:54:56 418

原创论文速看[2019-1-5]-Face Recognition: A Novel Multi-Level Taxonomy based Survey

一篇关于人脸识别中多任务的综述，submit to IET Biometricshttps://arxiv.org/ftp/arxiv/papers/1901/1901.00713.pdf摘要：In a world where security issues have been gaining growing importance, face recognition systems hav...

2019-01-05 13:55:20 464

原创 3D点云网络：PointNet:Deep Learning on Point Sets for 3D Classification and Segmentation

PointNet: Deep Learning on Point Sets for 3D Classification and SegmentationPointNet用于3D分类和分割的点云深度学习网络Stanford University， 2017CVPR斯坦福大学提出的直接对3D点云数据进行深度学习，是第一个直接处理点云数据的神经网络，同时还提出了PointNet++在2017年的n...

2018-12-15 15:44:00 2099

原创人脸识别-3D:Learning from Millions of 3D Scans for Large-scale 3D Face Recognition

Learning from Millions of 3D Scans for Large-scale 3D Face Recognition从百万张3D人脸图片中学习3D人脸识别2018 CVPR 西澳大利亚大学摘要原文译文

2018-11-20 21:56:17 3111 28

原创 Pytorch源码解读-torchvision.transforms

torchvision.transforms对于使用Pytorch的人肯定不陌生，我也用了Pytorch但是对transform却不是真正掌握，图片的预处理对网络的性能十分重要，所以打算仔细看看pytorch的源码。Transforms组成Transforms are common image transforms. They can be chained together using Co...

2018-11-16 20:18:09 1145 2

原创人脸识别 -关键点检测：Robust Facial Landmark Detection via a Fully-Convolutional Local-Global Context Network

Robust Facial Landmark Detection via a Fully-Convolutional Local-Global Context Network通过全卷积-局部全局语义网络实现鲁棒的关键点检测Technical University of Munich 慕尼黑科技大学摘要原文译文While fully-convolutional neura...

2018-11-13 14:38:40 1845

原创人脸识别-Pose(2):Dynamic Feature Learning for Partial Face Recognition

Dynamic Feature Learning for Partial Face Recognition使用动态特征学习进行部分人脸识别2018 CVPR 中科院自动化所智能感知与计算研究中心摘要原文译文Partial face recognition (PFR) in unconstrained environment is a very important tas...

2018-10-24 22:49:12 1173 1

原创 opencv实践之SVM

使用opencv中的SVM训练二分类器1 环境ubantu 16.04Python 3.7opencv3.4.3在Python环境下，sklearn包里的SVM更好用，但是因为项目需要，之后要将我的成果部署到C++中，所以选用了opencv来实现SVM。2 SVM3 Python实战3.1 任务我的目标是训练一个识别人脸是否戴眼镜的SVM分类器。这里我借助于facenet获取人...

2018-10-23 14:39:40 467

原创人脸识别-DeepVisage: Making Face Recognition Simple Yet With Powerful Generalization Skills

DeepVisage: Making Face Recognition Simple Yet With Powerful Generalization SkillsDeepVisage:利用强大的泛化能力使人脸识别更简单作者在摘要中总结了目前人脸识别算法的一些思路，但是目前流行的思路都需要额外的工作量或者数据，作者认为只有identity-label的数据就可以进行训练。文章提出用残差网络结构...

2018-10-19 16:29:58 415

原创人脸识别-Pose(1):DREAM:Pose-Robust Face Recognition via Deep Residual Equivariant Mapping

Pose-Robust Face Recognition via Deep Residual Equivariant Mapping通过深度残差等变映射实现抵抗姿态的人脸识别2018CVPR 香港中文大学、Sense Time作者github项目地址：DREAM摘要原文译文Face recognition achieves exceptional success th...

2018-10-17 21:37:29 1935

原创目标检测系列(4):MR-CNN和DeepBox

Object detection 之 MR-CNNObject detection via a multi-region &amp;amp; semantic segmentation-aware CNN modelObject detection 之 DeepBoxDeepBox: Learning Objectness with Convolutional Networks

2018-10-16 20:39:13 2543

原创目标检测系列(3):OverFeat和DeepMutliBox

Object detection 之 OverFeatObject detection 之 DeepMultiBoxScalable Object Detection using Deep Neural Networks利用深度神经网络实现可扩展的目标检测2014 CVPR，Google出品摘要文章指出之前的目标检测模型存在的问题：无法处理同类物体的多个实例cannot handle ...

2018-10-13 15:35:46 834

原创目标检测系列(2):SPP

Object detection 之 SPPSpatial Pyramid Pooling in Deep convolutional Networks for Visual Recognition用于视觉识别的卷积神经网络空间金字塔池化1 引入SPP的原因之前的CNN网络都要求输入图片的大小是固定的，比如AlexNet要求227x227、VGG要求224x224。这样的话，对于任意一张...

2018-10-08 20:23:21 895

原创目标检测系列(1):R-CNN

Object detectionRich feature hierarchies for accurate object detection and semantic segmentation用于目标检测和语义分割的多特征层级正式开始研究目标检测算法，从R-CNN开始，将‘一刀流’和‘二刀流’两种目标检测框架都了解研究一下。目的是：将目标检测算法中的先进技术和算法应用到人脸检测算法中，...

2018-10-08 11:13:32 256

原创 Pytorch源码解读-torch.utils.data

torch.utils.dataPytorch读取训练集需要用到torch.utils.data类，data类包括13个成员，主要用到的2个：class torch.utils.data.Dataset class torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=False, sampler=None, batch_s...

2018-09-25 11:04:08 2047

alfred_torres的博客