lama123's profile - activity

Thanks @kbarni, I will keep this in mind

I think you are right in case of pixel size, thanks So if we are using a resolution of 6MP (3000*2000) and we have max

I think you are right in case of pixel size, thanks So if we are using a resolution of 6MP (3000*2000) and we have max

5MP monochrome cameras give an accuracy of 2micrometre×2micrometre.Wont this be enough? How much transfer time do you t

Ok, Thanks for the info

Ok, Thanks for the info Amal

Camera for Gear Inspection Hello, We are planning to setup a camera system for gear inspection and defect detection. Th

The problem was with high learning rate. When the learning rate was changed from the default value of 0.01 to 0.0001 the

0% 10 20 30 40 50 60 70 80 90 100% |----|----|----|----|----|----|----|----|----|----| Training St

0% 10 20 30 40 50 60 70 80 90 100% |----|----|----|----|----|----|----|----|----|----| Training Star

you mean 0-9 right!! what do you mean by shuffling data? Currently testing with a reduced learning rate of 0.0001 instea

That is 97% accuracy..on mnist.. right So I wonder why we are not getting accuracy on NIST (link text

link text This is the link The change in size should be because of padding

link text This is the link

link text

and in mnist digits they are getting more than 98% accuracy with this architecture..so the doubt.

ok.. i will try this out The mnist database example in the tiny-dnn website does a -1 to 1 normalisation. This is the re

ok.. i will try this out

Training CNN for NIST digits using tiny-dnn I have been trying to train a CNN using tiny-dnn library for digit recogniti

OpenCV SVM performance poor compared to matlab ensemble Hello, I have been training a svm classifier for a 2 class forg

OpenCV SVM performance poor compared to matlab ensemble Hello, I have been training a svm classifier for a 2 class forg

//Set up SVM's parameters CvSVMParams params; params.svm_type = CvSVM::C_SVC; par

//Set up SVM's parameters CvSVMParams params; params.svm_type = CvSVM::C_SVC; params.kernel_t

//Set up SVM's parameters CvSVMParams params; params.svm_type = CvSVM::C_SVC; params.kernel_type = CvSVM:

Thanks for the clarification @berak Regards Amal

//Set up SVM's parameters CvSVMParams params; params.svm_type = CvSVM::C_SVC; params.kernel_type = CvSVM::RBF; params

Hello,

The values of p, nu and coef0 parameters are not getting updated in SVM::trainAuto training. It is always zero. The parameters gamma and cvalue are getting changed. Is this normal? Are the best values of these parameters usually zero?

The kernel type used is RBF and svm type is C_SVC

Also is there any way we can improve a two class classification other than making the parameter balanced = true in SVM::trainAuto training.

The project is forgery detection where we have to distinguish between forged and pristine images.

p, nu and coef0 parameters does not change during SVM::trainAuto training Hello, The values of p, nu and coef0 paramete

p, nu and coef0 parameters does not change during SVM::trainAuto training Hello, The values of p, nu and coef0 paramete

This is a new information, thanks. Just looked up transfer learning in google.

This is a new information, thanks. Just looked up in google about transfer learning.

true Are you getting high accuracy than svm even with just 3 layers ? I don't have much experience in deep networks. Lo

The main reasons for not using ANN are 1) Long training time 2) From experience other classifiers like svm gave better

The main reasons for not using ANN are 1) Long training time 2) From experience other classifiers like svm gave better

The main reasons for not using ANN are 1) Long training time 2) From experience other classifiers like svm gave better a

The main reasons for not using ANN are 1) Long training time 2) From experience other classifiers like svm gave better a

yep, that is the plan For 300 blocks per image and number of images equal to 500, the feature count will be 500*300=1,5

yep, that is the plan For 300 blocks per image and number of images equal to 500, the feature count will be 500*300=1,50

These are scrm features. A 2000x1000 image is divided into 32x 32 blocks with a stride of 16. Out of these blocks aroun

These are scrm features. A 2000x1000 image is divided into 32x 32 blocks with a stride of 16. Out of these blocks aroun

These are scrm features. A 2000x1000 image is divided into 32x 32 blocks with a stride of 16. Out of these blocks aroun

Training a classifier if feature size is greater than RAM size Hello, How can we train a classifier if the size of the

It would have been better if you provided an image for reference

Some tips you can use

1) Check for circular objects to detect ball. Take help of shape.

2) Look only in a small neighbourhood around the ball after the first frame while tracking the ball. This will reduce mistaking it with other objects.

Yes, i think you can use thresholding followed by findContours