The current section provides steps to use pretrained models:

1. Load tensorflow in R:

require(tensorflow) 
 

2. Assign the slim library from TensorFlow:

slimobj = tf$contrib$slim 

The slim library in TensorFlow is used to maintain complex neural network models in terms of definition, training, and evaluation.

3. Reset graph in TensorFlow:

tf$reset_default_graph() 
 

4. Define input images:

# Resizing the images 
input.img= tf$placeholder(tf$float32, shape(NULL, NULL, NULL, 3)) 
scaled.img = tf$image$resize_images(input.img, shape(224,224))

5. Redefine the VGG16 network:

# Define VGG16 network 
library(magrittr) 
VGG16.model<-function(slim, input.image){ 
  vgg16.network = slim$conv2d(input.image, 64, shape(3,3), scope='vgg_16/conv1/conv1_1') %>%  
    slim$conv2d(64, shape(3,3), scope='vgg_16/conv1/conv1_2')  %>% 
    slim$max_pool2d( shape(2, 2), scope='vgg_16/pool1')  %>% 
     
    slim$conv2d(128, shape(3,3), scope='vgg_16/conv2/conv2_1')  %>% 
    slim$conv2d(128, shape(3,3), scope='vgg_16/conv2/conv2_2')  %>% 
    slim$max_pool2d( shape(2, 2), scope='vgg_16/pool2')  %>% 
     
    slim$conv2d(256, shape(3,3), scope='vgg_16/conv3/conv3_1')  %>% 
    slim$conv2d(256, shape(3,3), scope='vgg_16/conv3/conv3_2')  %>% 
    slim$conv2d(256, shape(3,3), scope='vgg_16/conv3/conv3_3')  %>% 
    slim$max_pool2d(shape(2, 2), scope='vgg_16/pool3')  %>% 
     
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv4/conv4_1')  %>% 
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv4/conv4_2')  %>% 
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv4/conv4_3')  %>% 
    slim$max_pool2d(shape(2, 2), scope='vgg_16/pool4')  %>% 
     
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv5/conv5_1')  %>% 
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv5/conv5_2')  %>% 
    slim$conv2d(512, shape(3,3), scope='vgg_16/conv5/conv5_3')  %>% 
    slim$max_pool2d(shape(2, 2), scope='vgg_16/pool5')  %>% 
     
    slim$conv2d(4096, shape(7, 7), padding='VALID', scope='vgg_16/fc6')  %>% 
    slim$conv2d(4096, shape(1, 1), scope='vgg_16/fc7') %>%  
     
    slim$conv2d(1000, shape(1, 1), scope='vgg_16/fc8')  %>% 
    tf$squeeze(shape(1, 2), name='vgg_16/fc8/squeezed') 
  return(vgg16.network) 
} 
 

6. The preceding function defines the network architecture used for the VGG16 network. The network can be assigned using the following script:

vgg16.network<-VGG16.model(slim, input.image = scaled.img) 

7. Load the VGG16 weights vgg_16_2016_08_28.tar.gz downloaded in the Getting started section:

# Restore the weights 
restorer = tf$train$Saver() 
sess = tf$Session() 
restorer$restore(sess, 'vgg_16.ckpt')

8. Download a sample test image. Let's download an example image from a testImgURL location as shown in following script:

# Evaluating using VGG16 network 
testImgURL<-"http://farm4.static.flickr.com/3155/2591264041_273abea408.jpg" 
img.test<-tempfile() 
download.file(testImgURL,img.test, mode="wb") 
read.image <- readJPEG(img.test) 
# Clean-up the temp file 
file.remove(img.test)  

The preceding script downloads the following image from URL mention in variable testImgURL. The following is the downloaded image:

9. Determine the class using the VGG16 pretrained model:

## Evaluate  
size = dim(read.image) 
imgs = array(255*read.image, dim = c(1, size[1], size[2], size[3])) 
VGG16_eval = sess$run(vgg16.network, dict(images = imgs)) 
probs = exp(VGG16_eval)/sum(exp(VGG16_eval))  
 

The maximum probability achieved is 0.62 for class 672, which refers to the category--mountain bike, all-terrain bike, off-roader--in the VGG16 trained dataset.