Re: Some basic troubles

[Hossein Hasanpour]

Hello

1. I dont understand what you mean by that! if you happen to mean, how the learning process is carried out, it is done using back-propagation and gradient descent 

2.your dataset can be in leveldb/lmdb, hdf5 or even simply a folder containing your images and labels, this is explained in the layer catalog 

3.Caffe ships with two of these datasets, MNIST and CIFAR10 (soon cifar100). again there is a tutorial for both of these in  the documentation.

4.what you would basically do, is first train your network till you get a reasonable/desired accuracy, then you save your trained model(either using snapshots or using code), then 

you slightly alter your train_test prototxt file (your network definition file), and remove the data-layer where it reads data from the dataset.(the resulting file is called, deploy.protxt) 

then in code, you read your image, convert it and feed it to the network as your data, and then you will get your output. 

again this is as well explained and documented both in classify.cpp if you are doing c++, and classification.ipny in python(in python folder) which imho is way more informative and better at explaining and at the same time seeing the outcome in a step by step fasion.

On Tuesday, May 31, 2016 at 1:36:43 PM UTC+4:30, Rustam Mirzakhmedov wrote:

Hi everyone!

I just started to learn what neural networks is. So I've installed Caffe in Docker container, so what is my next steps?

1) How to learn network?

2) What dataset format should be?

3) Where I can get ready datasets or how make a new one?

4) How to insert a picture to get a result from learned Caffe?

Sorry for questions, I didn't find them in documentation. Good day everyone!