MATLAB Answers

Deep learning - AlexNet neural network re training.

13 views (last 30 days)
I'm using Alexnet to detect only one class of objects. I have re - trained alexnet with only one object class pencil over 100 pencil images. The training result is always 10020with no errors. When I'm using it for detection it gives 100 How to re-train this CNN accurately?

  0 Comments

Sign in to comment.

Accepted Answer

Kojiro Saito
Kojiro Saito on 27 Nov 2018

  3 Comments

ISURU SAMARAKOON
ISURU SAMARAKOON on 27 Nov 2018
Hi Kojiro,
Thank you very much for your reply. I refered to the web page but my problem is my training result is always 100% with no errors.I trained only 10 mages. I just want to know whether it is correct or a problem with the network. I've attached snap shot of training process.
(I trained using normal CPU)humantest1-1.jpg
Kojiro Saito
Kojiro Saito on 27 Nov 2018
I think it's because there's only one label "pencil". Any picture will be labeled as pencil. If you prepare images of other labels, the accuracy will be getting more accurate.
ISURU SAMARAKOON
ISURU SAMARAKOON on 27 Nov 2018
I will re-train the alexnet again with few more object classes and try the accuracy, anyway thank you so much for your reply.

Sign in to comment.

More Answers (1)

ugur arslan
ugur arslan on 25 Dec 2019
having same problem. did u getany progress?

  0 Comments

Sign in to comment.

Tags

Products


Release

R2018b

Translated by