(Day 19) TensorFlow Developer Certificate Part 2 - Convolutional Neural Networks in TensorFlow

Ivan Ivanov · January 20, 2024

Hello :) Today is Day 19!

A quick summary of today:

image image

Today I learned a bit more about the ImageGenerator

image

  • rescale=1./255 - pixel normalization
  • rotation_range=40 - rotates images 40 degrees
  • width_shift_range=0.2 - shifts images horizontally - image width
  • height_shift_range=0.2 - shifts images vertically within the range - image height
  • shear_range=0.2 - applies shear transformation
  • zoom_range=0.2 - zooms into images, up to a 20% zoom
  • horizontal_flip=True - flips images horizontally
  • fill_mode=’nearest’ Determines the strategy for filling in newly created (or missing) pixels after transformations; ‘nearest’ fills with the nearest existing pixel value

image

In the above case, overfitting can be corrected as shown in the picture below with Dropout(x)

image

Tomorrow’s part is Natural Language Processing, so it’s going to be fun because I don’t know anything about NLP.

That is all for today!

See you tomorrow :)

Original post in Korean