1 of 8

FUNDAMENTALS OF DEEP LEARNING

-SPANDAN SAR

2 of 8

MY LEARNINGS

  • Understanding the basic idea of vanilla neural networks and CNNs
  • Building and training models using Keras and TensorFlow
  • Build fairly accurate models with limited training data using image augmentation
  • Introduced to Google Colab and Keras documentation

3 of 8

PROJECT OBJECTIVE

  • Classifying 2 dog breeds using CNNs
  • Dog breeds- golden retriever and pug

MY LEARNINGS THROUGH THE PROJECT

  • Learnt about regularization through dropouts
  • Learnt how to play with the parameters and to structure the layers.
  • Learnt techniques to prevent overfitting and underfitting.

4 of 8

Classes classified

Image augmentation

5 of 8

MODEL

6 of 8

Accuracy and loss graphs

Train accuracy: 91%-95%

Validation accuracy:87%-90%

7 of 8

Prediction

400 X 400 pixels

28X28 pixels

8 of 8

THANK YOU