Is there any recent work on combining clustering approaches (k-means, or gaussian mixture or PGM) with deep learning for computer vision?

In particular I'm interested in if anyone has used the first few layers of a deep learning network as feature extractors in conjunction with clustering algorithms which have been engineered to induce things like translation and rotation invariance while preserving basic object structure?

Taking the max value out of the output of each feature and fitting them to a gaussian mixture is the easiest approach but I'm interested in seeing other ways you could structure the clustering algorithm. For example I'm interested in seeing how you might learn structure between features that includes position information.

  • $\begingroup$ My interpretation is that you want to take non-deep learning based clustering algorithms and use deep learning to improve on them in a collective way. If so, why not just use deep learning on feature data? There are several deep learning clustering algorithms. $\endgroup$ Jul 28 '20 at 14:43
  • $\begingroup$ Deep learning can be unreliable for consistent detection and sometimes produces completely wrong results without ability to interpret them. $\endgroup$ Jul 28 '20 at 15:54
  • $\begingroup$ Is your statement that "deep learning can be unreliable" a general one or is it in regards to results you have seen using DL-based clustering on your data? $\endgroup$ Jul 28 '20 at 16:18
  • $\begingroup$ My premise, wrong or right is that classic clustering algorithms are less prone to catastrophic failure that sometimes is seen neural networks so I'm interested in seeing how clustering algorithms could be used in conjunction with low level feature extractors. $\endgroup$ Jul 28 '20 at 17:43

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.