Lecture 6 | Testing Wild Images Against Mnist Trained Network

Поділитися
Вставка
  • Опубліковано 27 гру 2024

КОМЕНТАРІ • 9

  • @sawanisoman2505
    @sawanisoman2505 8 років тому

    Hello. I want to create an app, where user takes picture of a room number, that picture goes through neural net and checks with the data saved and gives the prediction. I am not sure, how to do that before the network. Please help me. Thank you

  • @hanxgen3
    @hanxgen3 8 років тому +2

    is this model can be used to another image object, for example to classify an image of fruit? and what if the testing and train image set use random height and weight rather than 28*28 pixels, will this model still work?

    • @Deeplearning4jSkymind
      @Deeplearning4jSkymind  8 років тому +1

      Yes this model could be used to classify fruit. Here is what you would change.
      Suppose you had 3 classes of fruit, apples, oranges and grapes, then you would set the output layer to have 3 neurons, one neuron for each class.
      You would then train the model on a collection of appropriate images.
      The code in this example scales the image to 28 * 28 and converts to greyscale.
      If your images where larger than that you could scale the input appropriately.
      Note that all images will be scaled to a consistent Height/Width.
      If you suspected that color information would lead to better result, then you would have to modify the input dataset and the ImageLoader to have 3 layers instead of one gray scale layer.
      Take a look at our examples,
      github.com/deeplearning4j/dl4j-examples
      Enjoy

    • @hanxgen3
      @hanxgen3 8 років тому +1

      Deeplearning4j - Skymind great explaination..thank you very much

  • @akilankabilan2929
    @akilankabilan2929 5 років тому

    Which software tool is used?

  • @sujanmaka2926
    @sujanmaka2926 8 років тому

    is this code suitable for hand written digit recognition? and the images must have white colored text and black background?

    • @Deeplearning4jSkymind
      @Deeplearning4jSkymind  8 років тому +1

      Hand written digit recognition is exactly what the code demonstrates.
      The challenge of background vs foreground is addressed briefly.
      The video is part of a series, so you can view the previous examples.
      It should be possible to train a network to recognize white on black as well as black on white. Check out the code, grab the dataset and flip half the images and see how it goes. Or automate the normalization to black on white or while on black by sampling the overall lightness or darkness of the image being tested and transform to match the format that the network was trained on.

  • @fatimahmath4819
    @fatimahmath4819 5 років тому +1

    sir can u just copying the code on command please

  • @Deeplearning4jSkymind
    @Deeplearning4jSkymind  8 років тому

    The code for this video is available.
    gist.github.com/tomthetrainer/218d4afc54be7928fc466441d68901c0