This repository contains the code and the results to classify images of cats and dogs.
Using Convolution Neural Network.
The CNN was developed using keras with Tensorflow as the backend.
Running on AWS p2.xlarge (Ohio) GPU instance.
Chen-Hsi (Sky) Huang (github.com/skyhuang1208)
An validation accuracy of 91.96% was achieved.
An score of 0.30379 was obtained from kaggle.
- Layer in: A batch (32) of 128x128 images
- Layer 1: Convolution2D(N of filters=32, Filter size= (3,3), activation='linear rectifier')
- Layer 1b: Maxpooling(pool size=(2,2))
- Layer 2: Convolution2D(N of filters=32, Filter size= (3,3), activation='linear rectifier')
- Layer 2b: Maxpooling(pool size=(2,2))
- Layer 3: Convolution2D(N of filters=64, Filter size= (3,3), activation='linear rectifier')
- Layer 3b: Maxpooling(pool size=(2,2))
- Layer 4: Dense(units=64, activation='linear rectifier')
- Layer 5: Dropout(rate=0.5)
- Layer 6: Dense(units=1, activation='sigmoid')
- Layer out: 1x1 matrix with probability of cat/dog.