TransferLearning

# Libraries used
library(keras)
library(EBImage)
library(tensorflow)

# Identify image 
pretrained <- application_resnet50(weights = 'imagenet')
setwd("~/Desktop")
img <- image_load("dog.jpg", target_size = c(224,224))
x <- image_to_array(img)
x <- array_reshape(x, c(1, dim(x)))
x <- imagenet_preprocess_input(x)
preds <- pretrained %>% predict(x)
imagenet_decode_predictions(preds, top = 5)[[1]]

# CIFAR10 data
data <- dataset_cifar10()
trainx <- data$train$x[1:2000,,,]
testx <- data$test$x[1:2000,,,]
trainy <- to_categorical(data$train$y[1:2000,], num_classes = 10)
testy <- to_categorical(data$test$y[1:2000,], num_classes = 10)

# Data preprocessing
x <- array(rep(0, 2000*224*224*3), dim = c(2000, 224, 224, 3))
for (i in 1:2000) {x[i,,,] <- resize(trainx[i,,,], 224, 224)}
trainx <- imagenet_preprocess_input(x)

x <- array(rep(0, 2000*224*224*3), dim = c(2000, 224, 224, 3))
for (i in 1:2000) {x[i,,,] <- resize(testx[i,,,], 224, 224)}
testx <- imagenet_preprocess_input(x)

# Model with RESNET50
pretrained <- application_resnet50(weights = 'imagenet',
                                   include_top = FALSE,
                                   input_shape = c(224, 224, 3))

model <- keras_model_sequential() %>% 
         pretrained %>% 
         layer_flatten() %>% 
         layer_dense(units = 256, activation = 'relu') %>% 
         layer_dense(units = 10, activation = 'softmax')
freeze_weights(pretrained)

# Compile
model %>% compile(loss = "categorical_crossentropy",
                  optimizer = 'adam',
                  metrics = 'accuracy')

# Fit model
history <- model %>% fit(trainx,
                         trainy,
                         epochs = 10,
                         batch_size = 10,
                         validation_split = 0.2)

# Evaluation & prediction
model %>% evaluate(testx, testy)
pred <- model %>% predict_classes(testx)
table(Predicted = pred, Actual = data$test$y[1:2000,])