I am currently working on a facial recognition project. I want to know that is htm a better choice rather than CNN because I don’t see a lot of work on facial recognition that uses htm. Please guide me.
I’m not extremely knowledgeable, but my two cents’ worth is: use a CNN. CNN’s are a more mature technology and highly specialized for that exact sort of task. But I’d wait to see if one of the experts responds before deciding.
Right, HTM hasn’t been established as competitive with the leading methods for image processing. I know there has been some work applying SP for MNIST digit recognition, but not facial recognition that I’m aware of.
HTM is really equipped to learn sequences, assuming the raw data type can be richly encoded into binary vectors (for conversion to SDR).
AFAIK there isn’t a well established HTM encoder for image data. This lack of encoders for certain data types limits HTM’s application scope as much as anything IMO.
For MNIST classification we achieve a classification rate of 97%.
With CNN you can get good results, but you do need any innovation!
For HTM you have more ideas and innovation to reach your goal. More challenges but I believe that HTM performance should be comparable to CNN!
It is only time question!
Benchmarking HTM on Yalefaces:
BTW, with CNN, most likely, you will get better results.