GMM and Bob tool
56 views
Skip to first unread message
Anita Garland
Apr 5, 2018, 12:30:59 AM4/5/18
to bob-devel
Hi,
I am wondering if you could provide answers to these questions.
What is the difference between
Could you please explain why the Tan & Triggs preprocessor is used with the GMM algorithm? And in that same note, why the DCT blocks feature extractor is used with GMM? Bob has numerous preprocessor and feature options available so is it just that those 2 provide the best results? Please pardon my ignorance as I'm new to Biometrics and GMM. If there is any online documentation on this, I would much appreciate it if you could direct me there.
Thank you kindly,
Anita
I am wondering if you could provide answers to these questions.
What is the difference between
bob.bio.gmm.algorithm.GMMRegular(**kwargs) and class bob.bio.gmm.algorithm.GMM? These are taken from https://www.idiap.ch/software/bob/docs/bob/bob.bio.gmm/master/implemented.html#bob.bio.gmm.algorithm.GMMRegular. And what exactly is kwargs? Is kwargs essentially the same as the argument list of the bob.bio.gmm.algorithm.GMM? Could you please explain why the Tan & Triggs preprocessor is used with the GMM algorithm? And in that same note, why the DCT blocks feature extractor is used with GMM? Bob has numerous preprocessor and feature options available so is it just that those 2 provide the best results? Please pardon my ignorance as I'm new to Biometrics and GMM. If there is any online documentation on this, I would much appreciate it if you could direct me there.
Thank you kindly,
Anita
Tiago Freitas Pereira
Apr 5, 2018, 3:15:56 AM4/5/18
to bob-...@googlegroups.com
Hi Anita,
The difference between bob.bio.algorithm.GMM and bob.bio.algorithm.GMMRegular is about the scoring function.
GMMRegular uses the classic log-likelihood ratio (LLR) between the client model and the UBM as scoring function (check it out
Reynolds, Douglas A., Thomas F. Quatieri, and Robert B. Dunn. "Speaker verification using adapted Gaussian mixture models." Digital signal processing 10.1-3 (2000): 19-41.)
The GMM uses an approximation of the LLR called Linear Scoring (check it out https://www.idiap.ch/software/bob/docs/bob/bob.learn.em/stable/guide.html#linear-scoring AND Glembek, Ondrej, et al. “Comparison of scoring methods used in speaker recognition with joint factor analysis.” Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on. IEEE, 2009.).
--
About **kwargs. Well, kwargs is a dictionary that corresponds to the keyword arguments of some function. Enabling it, will essentially allow your function to take an arbitrary number of keyword arguments.
Check it out this link for more information about its usage https://pythontips.com/2013/08/04/args-and-kwargs-in-python-explained/
--
Yes, you are right, bob.bio.face/bob.bio.base has numerous of preprocessors/extractors and you can run all sorts of permutations with them (you can even stack them).
And you definitely should do that by yourself to check it out :-)
The history about the triplet Tan&Triggs + DCT + GMM (no necessarily all together) that is suggested as a baseline is long.
The difference between bob.bio.algorithm.GMM and bob.bio.algorithm.GMMRegular is about the scoring function.
GMMRegular uses the classic log-likelihood ratio (LLR) between the client model and the UBM as scoring function (check it out
Reynolds, Douglas A., Thomas F. Quatieri, and Robert B. Dunn. "Speaker verification using adapted Gaussian mixture models." Digital signal processing 10.1-3 (2000): 19-41.)
The GMM uses an approximation of the LLR called Linear Scoring (check it out https://www.idiap.ch/software/bob/docs/bob/bob.learn.em/stable/guide.html#linear-scoring AND Glembek, Ondrej, et al. “Comparison of scoring methods used in speaker recognition with joint factor analysis.” Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on. IEEE, 2009.).
--
About **kwargs. Well, kwargs is a dictionary that corresponds to the keyword arguments of some function. Enabling it, will essentially allow your function to take an arbitrary number of keyword arguments.
Check it out this link for more information about its usage https://pythontips.com/2013/08/04/args-and-kwargs-in-python-explained/
--
Yes, you are right, bob.bio.face/bob.bio.base has numerous of preprocessors/extractors and you can run all sorts of permutations with them (you can even stack them).
And you definitely should do that by yourself to check it out :-)
The history about the triplet Tan&Triggs + DCT + GMM (no necessarily all together) that is suggested as a baseline is long.
This setup was sedimented along the history with our practical experiences and such practical experiences are reported and analysed in several papers.
Follow below some them in chronological order.
Have a special look in the last one (long survey full of experiments and analysis).
- Cardinaux, Fabien, Conrad Sanderson, and Samy Bengio. "Face verification using adapted generative models." Automatic Face and Gesture Recognition, 2004. Proceedings. Sixth IEEE International Conference on. IEEE, 2004.
- Cardinaux, Fabien, Conrad Sanderson, and Samy Bengio. "User authentication via adapted statistical models of face images." IEEE Transactions on Signal Processing 54.1 (2006): 361-373.
- Heusch, Guillaume, Fabien Cardinaux, and Sébastien Marcel. Lighting normalization algorithms for face verification. No. EPFL-REPORT-83268. IDIAP, 2005.
- McCool, Christopher, and Sébastien Marcel. "Parts-based face verification using local frequency bands." International Conference on Biometrics. Springer, Berlin, Heidelberg, 2009.
- Wallace, Roy, et al. "Inter-session variability modelling and joint factor analysis for face authentication." Biometrics (IJCB), 2011 International Joint Conference on. IEEE, 2011.
- McCool, Christopher, et al. "Session variability modelling for face authentication." IET biometrics 2.3 (2013): 117-129.
- de Freitas Pereira, Tiago, and Sébastien Marcel. "Periocular biometrics in mobile environment." Biometrics Theory, Applications and Systems (BTAS), 2015 IEEE 7th International Conference on. IEEE, 2015.
- Günther, Manuel, Laurent El Shafey, and Sébastien Marcel. "Face recognition in challenging environments: An experimental and reproducible research survey." Face recognition across the imaging spectrum. Springer, Cham, 2016. 247-280.
- Cardinaux, Fabien, Conrad Sanderson, and Samy Bengio. "Face verification using adapted generative models." Automatic Face and Gesture Recognition, 2004. Proceedings. Sixth IEEE International Conference on. IEEE, 2004.
- Cardinaux, Fabien, Conrad Sanderson, and Samy Bengio. "User authentication via adapted statistical models of face images." IEEE Transactions on Signal Processing 54.1 (2006): 361-373.
- Heusch, Guillaume, Fabien Cardinaux, and Sébastien Marcel. Lighting normalization algorithms for face verification. No. EPFL-REPORT-83268. IDIAP, 2005.
- McCool, Christopher, and Sébastien Marcel. "Parts-based face verification using local frequency bands." International Conference on Biometrics. Springer, Berlin, Heidelberg, 2009.
- Wallace, Roy, et al. "Inter-session variability modelling and joint factor analysis for face authentication." Biometrics (IJCB), 2011 International Joint Conference on. IEEE, 2011.
- McCool, Christopher, et al. "Session variability modelling for face authentication." IET biometrics 2.3 (2013): 117-129.
- de Freitas Pereira, Tiago, and Sébastien Marcel. "Periocular biometrics in mobile environment." Biometrics Theory, Applications and Systems (BTAS), 2015 IEEE 7th International Conference on. IEEE, 2015.
- Günther, Manuel, Laurent El Shafey, and Sébastien Marcel. "Face recognition in challenging environments: An experimental and reproducible research survey." Face recognition across the imaging spectrum. Springer, Cham, 2016. 247-280.
Cheers
--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/
---
You received this message because you are subscribed to the Google Groups "bob-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bob-devel+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
--
Tiago
Anita Garland
Apr 5, 2018, 10:41:36 PM4/5/18
to bob-devel
Thank you very much! This is very helpful.
Anita
Anita
--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/
---
You received this message because you are subscribed to the Google Groups "bob-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bob-devel+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
--Tiago
Reply all
Reply to author
Forward
0 new messages