![](https://img-blog.csdnimg.cn/20201014180756922.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
深度学习
ystsaan
这个作者很懒,什么都没留下…
展开
-
mnn在c++中的推理demo(mnn不支持5维输入,即暂时不支持3dcnn的推理)
准备工作1.使用cmake-gui在windows下编译mnn得到MNN.dll和MNN.lib(Release x64)2.opencv的dll,lib,头文件可以在https://github.com/opencv/opencv/releases/tag/4.3.0的opencv-4.3.0-vc14_vc15.exe直接安装后得到,不用编译(opencv_world430.dll,opencv_world430.lib)3.在vs2017中include mnn和opencv的所需要头文件(mn原创 2021-05-12 15:25:07 · 1050 阅读 · 1 评论 -
windows使用ssh连接服务器闲置过长后断开连接(Connection reset by server-ip port 22)
问题描述:最近下班前连接服务器训练模型开在那儿,一般第二天来都能继续训练不会断,但最近莫名其妙第二天来看的时候训练已经断了参考:https://blog.csdn.net/paincupid/article/details/91358564https://blog.csdn.net/u012222248/article/details/83060660上述第二篇博客写的使用的一种方法比较详细,第一篇博客写的方法中实际可以尝试的感觉也就只有两种方法一(在第二篇博客中写的比较详细)和方法四,且在方法四原创 2020-07-10 10:35:38 · 1450 阅读 · 0 评论 -
Counting Out Time: Class Agnostic Video Repetition Counting in the Wild个人笔记
参考https://mp.weixin.qq.com/s/olVWrq0Dca2Kulbc8xwplA由于上述公众号文章写的有些部分不够详细,比如repnet的transformer是什么,这里写一个自己读repnet和transformer论文的个人笔记repnet论文:http://openaccess.thecvf.com/content_CVPR_2020/papers/Dwibedi_Counting_Out_Time_Class_Agnostic_Video_Repetition_Coun原创 2020-07-02 16:16:22 · 643 阅读 · 1 评论 -
VidSage: Unsupervised Video Representational Learning with Graph Convolutional Networks个人论文笔记
模型主要从视觉和语义上编码一个视频:通过预训练的图像特征提取器提取视觉特征,再通过GCN变体来传播背景信息,获取视频语义3D CNN, LSTM, GCN(提取每帧人的关节作为关键点后产生的图像来预测动作)1.介绍通过GCN来对视频间的概念关联性进行建模:创建一个图Graph G,里面每个视频作为一个节点,然后基于视频间的语义相似性来形成节点间不同的连接;首先通过对比预训练好的图片特征提取器...原创 2020-04-15 13:47:34 · 981 阅读 · 0 评论 -
X3D: Expanding Architectures for Efficient Video Recognition个人论文笔记
https://zhuanlan.zhihu.com/p/129279351这篇X3D的解读写的比我早,写的挺好的,但有些细节没写上,所以由于强迫症写了这篇个人笔记X3D为一系列的高效视频分类网络(从2D图像分类网络不断的从各个方面(multiple network axes)进行扩张如帧频网络深度等等,每次只从一个方面进行扩张);渐进的前向扩张后是一个后向压缩从而达到一个固定复杂度的网络;X...原创 2020-04-17 21:29:49 · 2795 阅读 · 0 评论 -
SystemError: built-in function imread returned NULL without setting an error
背景:将含有字符串的list存入txt, ‘wb’和’rb’均为以二进制的方法写入和读取,所以在读取后的这个list中的元素都是为二进制的, 如b’/data/…/Out_0043.bmp’,前面有个b;这种目录给cv2.imread读取就会报错SystemError: returned NULL without setting an errorimport picklewith open...原创 2020-03-26 15:17:39 · 19862 阅读 · 0 评论 -
Tensorflow不能使用GPU的解决办法
转载于:https://blog.csdn.net/kudou1994/article/details/86735451 服务器在训练模型,另一边我在瞎胡乱搞不晓得咋个搞的,就不能使用GPU了。python3.6下from tensorflow.python.client import device_libprint(device_lib.list_local_devices())查...原创 2019-11-06 09:55:34 · 3922 阅读 · 0 评论 -
conda install出现段错误(segmentation fault)或是conda create创建环境出现段错误(segmentation fault)
https://github.com/conda/conda/issues/7815上面链接中的两种方法都用了就没问题了,具体是哪个操作解决问题还不清楚conda install pycosat==0.6.3conda clean -a原创 2019-11-03 15:02:40 · 7983 阅读 · 8 评论 -
指定多个GPU训练,GPU Memory Usage除了一个是满的其他都很低
https://github.com/keras-team/keras/issues/13057tensorflow 1.14.0有bug,使用其他版本的tensorflow原创 2019-11-03 13:08:19 · 847 阅读 · 0 评论 -
SlowFastNet(SlowFast) finetune(微调)
import syssys.path.append('.../SlowFast-master/slowfast/config/')sys.path.append('.../SlowFast-master/slowfast/models/')sys.path.append('.../SlowFast-master/slowfast/utils/')import slowfast.models...原创 2020-01-22 13:17:30 · 2680 阅读 · 3 评论 -
PyAV安装:段错误或者could not find libavdevice with pkg-config
PyAV安装时的错误直接使用conda install av -c conda-forge会出现段错误,所以把PyAV从github上下载下来,使用python setup.py install安装,会出现以下错误could not find libavdevice with pkg-config解决方案:参考:https://github.com/mikeboers/PyAV/issue...原创 2019-10-30 14:10:46 · 3687 阅读 · 0 评论 -
fatal:配置文件 /tmp/pip-p54sw5gt-build/.git/config 出错于第 1 行;git bad config line 1 in file
pip install ‘git+https://github.com/facebookresearch/fvcore’fatal: 配置文件 /tmp/pip-p54sw5gt-build/.git/config 出错于第 1 行Command “git clone -q https://github.com/facebookresearch/fvcore /tmp/pip-p54sw5gt...原创 2019-10-30 13:05:58 · 843 阅读 · 0 评论 -
Ubunu16.04安装GPU版本Pytorch(CUDA8.0 cudnn6.0 python2.7)
参考:https://blog.csdn.net/weixin_38708130/article/details/88895583conda install pytorch torchvision cudatoolkit=8.0 -c pytorch原创 2019-10-10 18:18:04 · 231 阅读 · 0 评论 -
tensorflow.python.framework.errors_impl.PermissionDeniedError
tf.summary.FileWriter原创 2019-09-25 10:56:01 · 2677 阅读 · 0 评论 -
tensorflow下的finetune(微调包括冻结某些层,添加或修改一些层,选择性读取权重)(恢复网络图时手动写下所有代码来创建图)
参考:https://zhuanlan.zhihu.com/p/53814653问题描述:网上搜到的csdn tensorflow的finetune的文章都不太容易令人理解,看了好几篇都无法在tensorflow下实现finetune;后来在bing上搜到了引用这篇参考网站的github上tensorflow finetune实现,非常的简洁易懂。该篇文章在tensorflow下实现finet...原创 2019-09-23 16:46:35 · 3474 阅读 · 1 评论 -
keras训练准确率和损失表现都非常好,预测时准确率和损失都非常差
问题简述:使用相同模型https://github.com/keras-team/keras/issues/6977上面github上的讨论主要说明了model.fit和model原创 2019-09-24 15:54:35 · 7685 阅读 · 0 评论 -
from scipy.misc import imread报错:ImportError: cannot import name imread
…原创 2019-07-31 08:28:32 · 314 阅读 · 0 评论 -
AttributeError: 'module' object has no attribute 'computation'
解决conda update dask原创 2019-07-30 16:52:05 · 329 阅读 · 0 评论 -
安装tensorflow及对应的cuda和cudnn
特简单。。。conda install tensorflow-gpu==1.3.0自动把cuda和cudnn都安装了,也都自动配置完了。。。直接略过网上各种教程。。。原创 2019-07-30 11:23:41 · 612 阅读 · 0 评论 -
ubuntu下的gdcm安装
gdcm安装特别简单。。。不用手动下载包 再make cmake。。。在terminal(终端)输入gdcmconv,它就会提示要使用某个命令安装gdcm包。。。输该命令就完事了。。。...原创 2019-07-30 11:19:17 · 2052 阅读 · 0 评论 -
labelme生成的json文件批量转化为label图片
修改C:\ProgramData\Anaconda3\Lib\site-packages\labelme\cli\json_to_dataset.py为import argparseimport base64import jsonimport osimport os.path as ospimport PIL.Imageimport yamlfrom labelme.log...原创 2019-03-06 14:39:54 · 11968 阅读 · 9 评论 -
py-faster-rcnn个人问题汇总
指令:make test问题:src/caffe/test/test_smooth_L1_loss_layer.cpp:11:35: fatal error: caffe/vision_layers.hpp: No such file or directory解决:src/caffe/test/test_smooth_L1_loss_layer.cpp重命名为test_smooth_L1_...原创 2019-03-10 15:36:29 · 561 阅读 · 0 评论 -
windows+anaconda+tensorflow
<div id="post_detail"><div class="post"> <h2> <a id="cb_post_title_url" href="https://www.cnblogs.com/HongjianChen/p/8385547.html">Windows10下用Anaconda3安装TensorFlow教程&l...转载 2019-04-29 16:05:08 · 168 阅读 · 0 评论 -
重组c#读取的图片一维数组为rgb图片或单通道灰度图图片
在keras+tensorflow框架下训练的权重可以在tf.keras.models.load_model调用,不必重新训练。同理,keras+tensorflow框架下的function,在这些function前加tf.同样可以实现原框架下的功能。import numpy as npimport tensorflow as tfimport cv2import osclass M...原创 2019-05-04 16:54:12 · 934 阅读 · 0 评论 -
c#调用python代码
https://blog.csdn.net/qq_42063091/article/details/82418630using System;using System.Collections.Generic;using System.Diagnostics;using System.Linq;using System.Text;using System.Threading.Tasks;...原创 2019-04-28 20:54:24 · 683 阅读 · 0 评论 -
import Keras/Tensorflow, python已停止工作解决方法
pip install h5py==2.8.0原创 2019-04-28 00:28:58 · 4031 阅读 · 7 评论 -
C#批量读取文件夹图片(图片宽度不为4的倍数所导致的偏差)
每次读取目录文件夹的一张图片并保存为数组。若图片为灰度图,保存的图片数组维度为原图的长乘宽;若图片为RGB三通图,保存的图片数组维度为原图的长乘3乘宽。具体维度可以在代码运行后从bmp.Stride和bmp.Height看出。using System;using System.Collections.Generic;using System.IO;using System.Linq;u...原创 2019-05-02 14:23:45 · 1043 阅读 · 0 评论 -
pythonnet c#调用并集成python代码
基本配置:win7Anaconda3-4.0.0-Windows-x86_64+python 3.5.1pythonnet-mastervs2017Anaconda3-4.0.0-Windows-x86_64.exehttps://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/(目前pythonnet不支持最新的python3.7)首先...原创 2019-04-29 23:16:20 · 5676 阅读 · 1 评论 -
keras+cntk后端在c#封装实现
神经网络在keras+cntk框架下训练得到权重cntk.combine(model.outputs).save(’/home/wxcnl/Downloads/cntk’)原创 2019-05-05 19:35:27 · 655 阅读 · 0 评论 -
One shot learning-Siamese Networks
One single image(neural networks don’t work well given one training example)(Such a small training set is really not enough to train a robust neural network)(If softmax, need to retrain every time ...原创 2019-05-21 10:43:51 · 657 阅读 · 0 评论 -
“ValueError : Tensor Tensor("predictions/Softmax:0", shape=(?, 2), dtype=float32) is not an element
原始问题及解决方案https://github.com/keras-team/keras/issues/2397#issuecomment-254919212问题描述:在keras+tensorflow框架下训练神经网络并得到权重h5文件。在之后需要调用的python代码中读取权重和图片并预测在C#多线程的子线程调用python代码时出现以下报错“ValueError : Tensor...原创 2019-05-07 09:42:05 · 2851 阅读 · 0 评论 -
使用pythonnet在C#进行多线程运行
问题描述:NET multi-threaded execution Python method, in Py. GIL can not run解决方案:https://github.com/pythonnet/pythonnet/issues/109具体解决步骤,下载pythonnet-master,解压后打开pythonnet.sln,点击Python.Runtime项目目录下的python...原创 2019-05-06 19:48:01 · 1176 阅读 · 1 评论 -
ImportError : Something is wrong with the numpy installation.
问题:ImportError : Something is wrong with the numpy installation. While importing we detected an older version of numpy解决方法:pip uninstall numpyconda uninstall numpypip install numpy...原创 2019-04-30 12:38:29 · 3585 阅读 · 0 评论