Deep-Learning-Tutorial

Deep Learning Pipeline from start to finish

Tutorial Description

In this Tutorial I want to give a clear view of the How Deep Learning works from collecting Dataset to get actual results and predictions, in this example we create a Deep Learning Based algorithm that can recognize two movies

Ernest Celestine Movie	Toy Story 3 Movie

Steps

Step 1

Importing data is the most important thing where you only have to create a folder called Dataset in a structure as follows:

• Dataset (Folder)
  • Class 1 (Folder)
  • Class 2 (Folder)

Example (File Tree):

./Dataset:

• ./Dataset/0

  • image1.png
  • image2.png
  • .. ect

• ./Dataset/1

  • image1.png
  • image2.png
  • .. ect

Note :

Name of the folder is the label of that class

Step 2

Create Your Model Layers

Example (Model used in this tutorial)

Model Type : "sequential"

Layer (type)	Output Shape	Param #
conv2d (Conv2D)	(None, 268, 478, 8)	224
max_pooling2d (MaxPooling2D)	(None, 134, 239, 8)	0
conv2d_1 (Conv2D)	(None, 134, 239, 8)	1168
max_pooling2d_1 (MaxPooling2D)	(None, 66, 118, 16)	0
flatten (Flatten)	(None, 124608)	0
dense (Dense)	(None, 100)	12460900
dense_1 (Dense)	(None, 2)	202

• Total params: 12,462,494
• Trainable params: 12,462,494
• Non-trainable params: 0

Step 3

Plot your results of accuracy and lost

Example (My Results)

Note : This is My Results but note that you can do better if you found the Right Layers

• Accuracy Results

• Loss Results

Detailed Review about the code

In this code you will find:

• Function create_dataset that reads your Dataset Folder Depending on how much class on your folder

  • Output:
    • your images as an array
    • array of labels (Depending on your folder name )

• Function Shuffle in unison shuffles two arrays in the same manner so we don't lose track of our labels

• Model layers

• Saving Model history as Dictionary

• Plotting Dictionary Values: to show the progress each epoch