naureen20/image-classification-using-transfer-learning-and-optimal-learning-rate-estimation

Classifying Images of 10 categories of animals taken from Kaggle Database using VGG16, ResNet-50 and imagenet. Learning Rates for respective models is Estimated from Epoch Vs Learning Rate graph obtained using Learning rate finder function

Image-Classification-using-Transfer-Learning-and-optimal-Learning-Rate-Estimation

Classifying Images of 10 categories of animals, using VGG16, ResNet-50 and imagenet. Images have been collected from Google and launched in Kaggle Data sets. Learning Rates for respective models is Estimated from Epoch Vs Learning Rate graph obtained using Learning rate finder function. Optimal learning rate for VGG16 model is found to be 1e-4 and for ResNet50 as 5e-5. VGG16 model gave a training accuracy of 97.040 % and validation accuracy of 96.210 %.ResNet50 gave a train accuracy=90.222 %,Val accuracy=90.170 %.

Image Preprocessing Includes Resizing, Reshaping, One-Hot vectorizing categories(animal names), Changing label names from Spanish to English, train-validation split and Data Augmentation

Creating CNN Models with Transfer Learning Includes creating VGG16 and ResNet50 models, obtaining a graph of Learning Rate vs epochs using the defined Defining LEARNING RATE FINDER function

The above image gives optimal learning rate for ResNet50

The above image gives optimal learning rate for VGG16. >It is Observed that VGG16 gave better accuracy than ResNet50