Is there a good package for image classification except for keras in R?

Machine Learning and Modeling
1

I am a beginner in R and trying to work with image recognition as a project. I have images of dogs and cats and wanted to know if there is a good reference of source code or a package that will work as a neural network for image classification

1 Like
2

It would be easier to help you if you could explain why you don't want to use keras. Anyway, AFAIK, you have 3 options:

  1. Use tensorflow directly, without the keras API.
  2. Use the OpenCV R wrapper, but: a) it's very experimental and it might crash R and b) it's likely that the best method will be based on a convolutional neural network anyway, thus you would be still using keras or tensorflow under some wrapper, and miss the opportunity to learn how to use them at their fullest
  3. Write the code yourself. How accurate you want the classifier to be, and where are you going to get the photos from? Depending on your accuracy target and the complexity of your dataset, even simple logistic regression or a SVM may get you to > 70% accuracy, if you just need to discriminate between dogs and cats. Writing a CNN/MLP from scratch (including backpropagation) for this task is fairly easy in Python, thanks to NumPy. I've never done it from scratch in R, so I'm not sure if it would be faster/slower.
3

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.