Dataset for 3D CNN

Hello everyone

I would like to create a classification model with the following characteristics:

  • 4 different classification classes
  • Each class directory contains several thousand measurement point directories
  • Each measuring point directory contains 10 png files of the same size (images that follow each other
    like a movie).
    I have all the data on my disk

How do I load the data into a dataset ?
Should I use a 3D CNN? LSTM?
How do I do it ?

Thanks for your help!

Hi @Ecostate, Please try to arrange the images in a way that each sub directory contains images corresponding to one class like


so that you can load the images easily through tf.keras.utils.image_dataset_from_directory( ).

You can try with different models containing different layers like conv2d, lstm, conv3d and select the model which gives best results. Thank You.