c++ - Caffe classifocation.cpp always returns 100% probability -

-

i'm trying use caffe c++ classification example (here code) classify image handwritten digit (i train model on mnist database), returns probabilities 

[0, 0, 0, 1.000, 0, 0, 0, 0, 0]  (1.000 can on different position)

even if image has no number on it. think should like

[0.01, 0.043, ... 0.9834, ... ]

also, example '9', it's predicts wrong number.
1 thing change in classification.cpp i'm using cpu 

//#ifdef cpu_only       caffe::set_mode(caffe::cpu); // <----- cpu //#else //  caffe::set_mode(caffe::gpu); //#endif

this how deploy.prototxt looks like

name: "lenet" layer {   name: "data"   type: "imagedata"   top: "data"   top: "label"   image_data_param {     source: "d:\\caffe-windows\\examples\\mnist\\test\\file_list.txt"   } } layer {   name: "conv1"   type: "convolution"   bottom: "data"   top: "conv1"   param {     lr_mult: 1   }   param {     lr_mult: 2   }   convolution_param {     num_output: 20     kernel_size: 5     stride: 1     weight_filler {       type: "xavier"     }     bias_filler {       type: "constant"     }   } } layer {   name: "pool1"   type: "pooling"   bottom: "conv1"   top: "pool1"   pooling_param {     pool: max     kernel_size: 2     stride: 2   } } layer {   name: "conv2"   type: "convolution"   bottom: "pool1"   top: "conv2"   param {     lr_mult: 1   }   param {     lr_mult: 2   }   convolution_param {     num_output: 50     kernel_size: 5     stride: 1     weight_filler {       type: "xavier"     }     bias_filler {       type: "constant"     }   } } layer {   name: "pool2"   type: "pooling"   bottom: "conv2"   top: "pool2"   pooling_param {     pool: max     kernel_size: 2     stride: 2   } } layer {   name: "ip1"   type: "innerproduct"   bottom: "pool2"   top: "ip1"   param {     lr_mult: 1   }   param {     lr_mult: 2   }   inner_product_param {     num_output: 500     weight_filler {       type: "xavier"     }     bias_filler {       type: "constant"     }   } } layer {   name: "relu1"   type: "relu"   bottom: "ip1"   top: "ip1" } layer {   name: "ip2"   type: "innerproduct"   bottom: "ip1"   top: "ip2"   param {     lr_mult: 1   }   param {     lr_mult: 2   }   inner_product_param {     num_output: 10     weight_filler {       type: "xavier"     }     bias_filler {       type: "constant"     }   } } layer {   name: "loss"   type: "softmax"   bottom: "ip2"   top: "loss" }

file_list.txt 

d:\caffe-windows\examples\mnist\test\test1.jpg 0

and tests1.jpg this

enter image description here

(black&white 28*28 image saved in paint, have tried different sizes doesn't matter, preprocces() resizes anyway)

to train network use this tutorial, here prototxt

so why predicts wrong digits , alway 100% probability?

(i'm using windows 7, vs13)

in "imagedata" layer, should normalize test1.jpg data [0, 255] [0, 1] "scale" keep consistency of preprocess manner between training , test following: 

image_data_param {     source: "d:\\caffe-windows\\examples\\mnist\\test\\file_list.txt"     scale: 0.00390625   }

Comments

Post a Comment

Popular posts from this blog

matlab - error with cyclic autocorrelation function -

-
Image
i'm trying code cyclic autocorrelation function in matlab follows: t=0:(n-1); t=t*te; i_alpha=0; tau=-n2*te:te:n2*te; alpha=-1/2:1/n:1/2; ryy_cl=zeros(length(alpha),length(tau)); alpha=alpha/te ind_tau=0; i_alpha=i_alpha+1; k=tau ind_tau=ind_tau+1; if k>0 % ryy_cl(i_alpha,ind_tau)=1/length(sig_bin_syl_mod(1:end- k))*sum((sig_bin_syl_mod(1:end-k)).*sig_bin_syl_mod(k+1:end)).*exp(1i*2*pi*alpha*t(1+k:length(sig_bin_syl_mod(1:end)))); ryy_cl(i_alpha,ind_tau)=(1/n)*sum((sig_bin_syl_mod(1:end-k)).*sig_bin_syl_mod(k+1:end)).*exp(-1i*2*pi*alpha*t(1:end-k)); else ryy_cl(i_alpha,ind_tau)=(1/n)*sum((sig_bin_syl_mod(1-k:end)).*sig_bin_syl_mod(1:end+k)).*exp(-1i*2*pi*alpha*t(1-k:end)); end end end i'm getting errors, , doesn't show expecting. according below formulae, how can fix it? the code provided incomplete, difficult confidently reproduce error. in particular, haven't de...
Read more

django - (fields.E300) Field defines a relation with model 'AbstractEmailUser' which is either not installed, or is abstract -

-
i trying create customuser django project email username , add radio button truck , company. in registration process email-id registered per truck or company. mentioned radio button 'tag' , add manytomany field tag in emailuser model. when makemigrations, raising error : (fields.e300) field defines relation model 'abstractemailuser' either not installed, or abstract. i quite new django , not sure whether have created correct code wants. please me solving this. here code, models.py: import django django.contrib.auth.models import ( abstractbaseuser, baseusermanager, permissionsmixin) django.core.mail import send_mail django.db import models django.utils import timezone django.utils.translation import ugettext_lazy _ class emailusermanager(baseusermanager): """custom manager emailuser.""" def _create_user(self, email, password, is_staff, is_superuser, **extra_fields): """create , save emailus...
Read more

c# - What is a good .Net RefEdit control to use with ExcelDna? -

-
i using exceldna , c# build suite of tools. part of forms i'd want user select cells/ranges workbook. i've got application.inputbox route working...but doesn't cool... i haven't read many things excel's refedit control, , see many have in past written workarounds, in (shudder) vba or vb.net. however, none of these seem recent... best approach/control refedit functionality on form run via exceldna (if applicable)? the refedit sample project of github explores different ways of dealing threading when showing form excel add-in, main issue making refedit control .net add-in. check page links various other implementations: https://github.com/excel-dna/exceldna/wiki/links-about-refedit
Read more