Build Online Image Classification Service with Shiny and MXNetR

[This article was first published on DMLC(distributed machine learning common), and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Early this week, Google announced its Cloud Vision API, which can detect the content of an image.

With the power of R and MXNet, you can try something very similar on your own laptop: an image classification shiny app. Thanks to the powerful shiny framework, it is implemented with no more than 150 lines of R code.


Installing mxnet package

Due to various reasons, mxnet package can't get on cran, but we try our best to make installation process easy.

For Windows and Mac users, you can install CPU-version of mxnet in R directly using the following code:

install.packages("drat", repos="")

If you want to use the power of your GPU or you are a Linux hacker, please follow the link.

Run the shiny app

Besides the mxnet, you will also need shiny for the web framework and imager for image preprocessing. Both of them are on CRAN, so you can install them easily.

install.packages("shiny", repos="")
install.packages("imager", repos="")

The hardest part has been done if you finish all the installation. Let's run the app and have fun! You can clone the repo or just use the line of code below in R.


For the first time, it will take some time to download a pre-trained Inception-BatchNorm Network (you can know more details on network architecture from the paper). And then you can use your local figures or a url containing figure. Personally I think the result is quite good.


Code behind the app

You can find all the code from the repo. Just like other shiny apps, we have ui.R and server.R. The ui.R is quite straightforward, we just define a sidebarPanel and a mainPanel.

Let's look into the server.R. All the web-related things are nicely handled by shiny. Besides that, there are 3 chunks of code.

First, we load the pre-trained model:

model <<- mx.model.load("Inception/Inception_BN", iteration = 39)
synsets <<- readLines("Inception/synset.txt")
mean.img <<- as.array(mx.nd.load("Inception/mean_224.nd")[["mean_img"]])

Then we defined a function to preprocess figures:

preproc.image <- function(im, mean.image) {
  # crop the image
  shape <- dim(im)
  short.edge <- min(shape[1:2])
  yy <- floor((shape[1] - short.edge) / 2) + 1
  yend <- yy + short.edge - 1
  xx <- floor((shape[2] - short.edge) / 2) + 1
  xend <- xx + short.edge - 1
  croped <- im[yy:yend, xx:xend,,]
  # resize to 224 x 224, needed by input of the model.
  resized <- resize(croped, 224, 224)
  # convert to array (x, y, channel)
  arr <- as.array(resized)
  dim(arr) = c(224, 224, 3)
  # substract the mean
  normed <- arr - mean.img
  # Reshape to format needed by mxnet (width, height, channel, num)
  dim(normed) <- c(224, 224, 3, 1)

Last, we read the figure and make prediction:

im <- load.image(src)
normed <- preproc.image(im, mean.img)
prob <- predict(model, X = normed)
max.idx <- order(prob[,1], decreasing = TRUE)[1:5]
result <- synsets[max.idx]

If you met any problem, please just open an issure. Any PR will be truly appreciated! If you find really interesting results, share it on Twitter with #MXnet!


I know it is not a Chihuahua, but very close.


A photo in my phone and just try yours!

To leave a comment for the author, please follow the link and comment on their blog: DMLC(distributed machine learning common). offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Never miss an update!
Subscribe to R-bloggers to receive
e-mails with the latest R posts.
(You will not see this message again.)

Click here to close (This popup will not appear again)