2016年03月_vwenyu-L

12月 11月 10月 09月 08月 07月 06月 05月 04月 03月 02月

原创 caffe --- mistakes

1. Check failed: img_height == data_mean_.height() (224 vs. 256)layer { name: "data" type: "ImageData" top: "data" top: "label" include { phase: TRAIN } transform_param {

2016-03-30 21:03:07 902 2

原创 about image ranking --- papers

Cross-domain Image Retrieval with a Dual Attribute-aware Ranking Network

2016-03-30 16:25:43 1016

原创 linux-vim

vi 包含两种命令模式 vi (较短并且不回显)+ ex(: xxx)-----------------光标的移动---------尽可能用最少的次数移动到正确的位置-----------1. h j k l ( 左下上右) 2. w e b(向前一个单词开始出单词结束处向后) W E B (忽略标点) 在前面可以加上重复次数3. H M

2016-03-28 22:24:18 393

原创 implement of deep neural network --- python

import numpy as npimport randomdef sigmoid(z): return 1.0/(1.0 + np.exp(-z))def sigmoid_prime(z): return sigmoid(z)*(1.0-sigmoid(z))class Net(object): def __init__(self,

2016-03-26 21:22:23 519

原创 python-numpy-00

ndarray.flatten('C' or 'F') plt.hist( ndarr , bins=xxx, normed = 1, )

2016-03-26 16:12:57 531

原创 linux --- shell command

-----------------grep----------------------1. grep -F(fixed-string) xxx -r -R(path name) xxx/ 2. grep -v(invert-match, non-match) -f(file) f1 f2 # output f2 but without the line in f1 # -f FI

2016-03-25 13:05:45 428

原创 python-language/tips/

------------0----glob-------------xx = glob.glob('xxx') [] * . return listyy = glob.glob('xxx') return generator---------1---------

2016-03-24 19:13:18 362

原创 digit image process --- python-skimage

1. Anaconda,conda list, conda install xxx, conda update xxx--------------------0------------------from skimage import ioimg = io.imread(xxx)io.imshow(img)

2016-03-24 16:12:30 864

原创 BN: Accelerating Deep Network Training by Reducing Internal Covariate Shift

2016-03-23 16:48:44 309

原创 network in network---NIN&mlpConv

mlpConv:add some 1*1 conv(ReLU) after the conv------------------------------global average pooling without fc layer

2016-03-23 09:37:55 1454

原创 going deeper with convolution---googlenet

sss

2016-03-22 14:08:25 647

原创 very deep convolutional networks for large-scale image recognition---vggnet

这篇论文提到的模型基本结构和alexnet特别类似：vgg: 提出的是一个网络模型，作者试验了不同参数层从11到19 （weight layers）值得关注的几点：1.为什么小的卷积核和比大的卷积核的参数更少比如三个3X3的卷积核的参数与一个7X7的卷积核在输入中的reception field一样大，但是3x3的参数个数 3x(3x3xInputxOutput）

2016-03-21 11:36:36 1195

原创 source of deep learning and ml

------------------0------------------------------#/usr/bin/bahIMAGE_ROOT=xxxrm -rf train.txtfind IMAGE_ROOT/image/ -name *xx.jpg | cut -d '/' -f3 | sed "s/$/ 1/" >> train.txtfine IMAGE_ROOT/

2016-03-10 15:58:19 1022

原创 usage-of-caffe

1. training stage ----- log -----xxx.sh!/bin/bashcaff_root=xxxsolver_file = xxxcaffe_file = caffe_root+'built/tools/caffe$caffe train --solver $solver_file | tee $log_file----------------

2016-03-10 09:19:16 914

原创 git-github

------------------------git---------------1. git clone XXX2. git init3. git add xxx4. git commit -m 'xxx'5. git push origin master--------------------------------------------------------

2016-03-09 15:26:50 400