-2

I have question about design of Neural Net.

I'm quite new to topic, but I have solid intuition about it. I know I should use convolutional neural network for it. But what existing design of net may I use? I want to classify some object on picture, for instance road signs. Where should I look?

Thank you!

Július Marko
  • 1,196
  • 5
  • 15
  • 37

2 Answers2

2

Yes, fine-tuned convolutional neural network is a cutting edge architecture for such task.

Usually, CNN is composed of successive convolutional and pooling layers and has some fully-connected layers on the output.

I'd strongly recommend you to check out this link for detailed explanation.

Dmitriy Danevskiy
  • 3,119
  • 1
  • 11
  • 15
  • even I don't have such mathematical background for it? – Július Marko Sep 28 '16 at 18:39
  • 1
    You don't really need to be proficient at math to use deep learning algorithms. Definitely, some basic math skills are required (calculus basics, vectors, matrices, optimization) but it takes only a couple of weeks (or even days) to obtain a level sufficient for using deep learning even if you're started from scratch. – Dmitriy Danevskiy Sep 28 '16 at 18:57
-1

CNN is OK.

I wouldn't use tensorflow. I suggest to look at the framework caffe. Sorry for my bad english.

Gergely Reményi
  • 34
  • 3
  • Why you would not to use tensorflow? Do you have reason for it? Or someone other has something against TensorFlow? I want to do it with TF, but I'm newbie, so any good tips are welcome – Július Marko Sep 29 '16 at 20:42
  • http://deeplearning4j.org/compare-dl4j-torch7-pylearn.html Look at this site. I think, there stand nearby facts. – Gergely Reményi Sep 29 '16 at 21:49