Model used: Inception V3
Layers added: MaxPooling, Dropout, Convolution, Flattening and Full connection Epochs: 3 Batch size: 64 Steps per epoch: 10 Validation: 10 validation steps after each of three epochs
Accuracy on test set (without additional Convolution2D layer): ~83% After I changed MaxPooling to shrink layers in 2 times instead of 8 times and applied Convolution2D, the accuracy increased as following Accuracy on test set: ~95%
Images of cats and dogs
$ git clone
$ pip3 install requirements.txt$ python3 transfer-learning.py