I am currently working on a image classification application using deep learning algorithms (either by using GIST features or CNN). I need help in understanding the below queries.

  1. I have extracted the GIST features of an image (Reference Link). These extracted features will be given as input to deep learning algorithm to classify the images. Is there a way to visualize the extracted features on top of the image?

  2. CNN or GIST, Which is better for image classification? Is GIST outdated when compared to CNN?

Thank you, KK


Given that code is trivial for both (GIST + Network and Raw Pixel + Network), you can try three approaches for a given project.

  1. GIST + Dense layers (GIST is not space-distributed)
  2. Raw Pixels + CNN + Dense Layers
  3. Raw pixels + CNN + Dense + input layer 2 (GIST) + Dense

For some projects, GIST can help since it is an abstract feature that CNN might or might not learn.

EDIT: This paper compares GIST and CNN


Is there a way to visualize the extracted features on top of the image?

This can be done with an attention layer in approach 3 (CNN + GIST).

CNN provides spacial distribution (Required for visualization) and dense layer that merges CNN's output with GIST can be used with an attention layer.

Paper for visualization

  • $\begingroup$ Added some details on visualization $\endgroup$ Mar 20 '19 at 9:55
  • $\begingroup$ Thank you for the information. Really helped me understanding the concepts. Will check the feasibility of implementation and update you here. $\endgroup$
    – deepguy
    Mar 21 '19 at 3:04
  • $\begingroup$ Is there a way I can visualize the extracted GIST features? The link you have given seems to be for the CNN (Attention based). $\endgroup$
    – deepguy
    May 14 '19 at 8:34

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.