I searched through the internet but couldn't find a reliable article that answers this question.

Can we use Autoencoders for unsupervised CNN feature learning of unlabeled images like the below enter image description here and use the encoder part of the Auto-encoder for Transfer learning of few labeled images from the dataset? as shown below. enter image description here

I believe this will reduce the labeling work and increase the accuracy of a model.

However, I have concerns like the more cost in computing, failing to learn all required features, etc..

Please let me know if any employed this method in large scale learning such as image-net.

PS: Pardon if it is Trivial or Vague as I am new to the field of AI and computer vision.

  • $\begingroup$ Are you asking if there is a way of training an encoder to label images (or predict the class of an image), so that this can be used to train another NN in a supervised way, where the labels are provided by this trained encoder? $\endgroup$ – nbro May 11 '19 at 20:48
  • $\begingroup$ @nbro, No, I just want to use the encoder part of the Auto encoder as the feature extractor in the image recognition neural network. As it would have already learned some low-midlevel features of images. $\endgroup$ – Tamilarasu Ulaganathan May 17 '19 at 18:01

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.