Import CIFAR - 10 Dataset for R

Question

I am trying to load a CIFAR-10 image dataset; http://www.cs.toronto.edu/~kriz/cifar.html in R, but I can't seem to extract the files. I tried all three formats .bin, .mat and python. Can someone help with some suggestions on how to extract them?

Thanks a lot, Will

Answer 1

As in any case, I would say that the easiest way, as a rule, is contrailing on someone else's zeal. For this case, this means that you need to look for someone else who has already transformed it. A quick Google search reveals this site (which contains an R image data file) is an excellent candidate for this method.

Alternatively, if you want to directly use the CIFAR-10 data, here the script I just quickly created data from a binary file that Alex contacted with the source page for cifar-10 :

# Read binary file and convert to integer vectors # [Necessary because reading directly as integer() # reads first bit as signed otherwise] # # File format is 10000 records following the pattern: # [label x 1][red x 1024][green x 1024][blue x 1024] # NOT broken into rows, so need to be careful with "size" and "n" # # (See http://www.cs.toronto.edu/~kriz/cifar.html) labels <- read.table("cifar-10-batches-bin/batches.meta.txt") images.rgb <- list() images.lab <- list() num.images = 10000 # Set to 10000 to retrieve all images per file to memory # Cycle through all 5 binary files for (f in 1:5) { to.read <- file(paste("cifar-10-batches-bin/data_batch_", f, ".bin", sep=""), "rb") for(i in 1:num.images) { l <- readBin(to.read, integer(), size=1, n=1, endian="big") r <- as.integer(readBin(to.read, raw(), size=1, n=1024, endian="big")) g <- as.integer(readBin(to.read, raw(), size=1, n=1024, endian="big")) b <- as.integer(readBin(to.read, raw(), size=1, n=1024, endian="big")) index <- num.images * (f-1) + i images.rgb[[index]] = data.frame(r, g, b) images.lab[[index]] = l+1 } close(to.read) remove(l,r,g,b,f,i,index, to.read) } # function to run sanity check on photos & labels import drawImage <- function(index) { # Testing the parsing: Convert each color layer into a matrix, # combine into an rgb object, and display as a plot img <- images.rgb[[index]] img.r.mat <- matrix(img$r, ncol=32, byrow = TRUE) img.g.mat <- matrix(img$g, ncol=32, byrow = TRUE) img.b.mat <- matrix(img$b, ncol=32, byrow = TRUE) img.col.mat <- rgb(img.r.mat, img.g.mat, img.b.mat, maxColorValue = 255) dim(img.col.mat) <- dim(img.r.mat) # Plot and output label library(grid) grid.raster(img.col.mat, interpolate=FALSE) # clean up remove(img, img.r.mat, img.g.mat, img.b.mat, img.col.mat) labels[[1]][images.lab[[index]]] } drawImage(sample(1:(num.images*5), size=1)) 

Hope this helps!