I am new to AI and NN. I've started learning using Geron's book on Tensorflow.

My first project ("Smart Shelf") is to determine which items in a store have been purchased and need refilled. The store camera periodically takes pictures of the tops of items on store shelves. To start, we have only 5 distinct products.

We have created ~250 handwritten images of product-labels that cover these 5 distinct products. So far, the training results are way below our expectation.

I am thinking to augment the training data and see whether it would make any difference.

I have thought about the following strategies:

  1. Train the model again using grayscale images. https://stackoverflow.com/questions/45320545/impact-of-converting-image-to-grayscale/45321001
  2. Invert images, translate them horizontally or vertically https://www.tensorflow.org/tutorials/images/data_augmentation, https://nanonets.com/blog/data-augmentation-how-to-use-deep-learning-when-you-have-limited-data-part-2/

Which of the above will yield better results and why? I am curious. Thanks for any help. I feel that I know various data augmentation techniques, but not sure how and why to apply them.

It seems this is a popular question, as learned from Choosing Data Augmentation smartly for different application etc.

  • 1
    $\begingroup$ Hi! Sorry but I'm not understand about this part "~250 handwritten images that cover these 5 distinct products", can you edit your question and give more information or examples? $\endgroup$
    – malioboro
    Feb 8 at 9:35
  • $\begingroup$ Sure Malioboro. Those handwritten images are of data labels. $\endgroup$ Feb 9 at 20:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.