bioinspired.Retina: log-polar representation vs retinal with log sampling

[breznak]

Detailed description

The advantage of log-pol coordinates in computer vision is that

By using this transformation, both rotations and scaling between two images can be represented by translations.

https://sthoduka.github.io/imreg_fmt/docs/log-polar-transform/
This means those will have similar representations (for use in neural net)

bioinspired.Retina has an option for useRetinalLogSampling, eg as shown at:
http://users.isr.ist.utl.pt/~alex/Projects/TemplateTracking/logpolar.htm

I am wondering:

• any advantages of logpolar over retinal log-sampling?
• is the log-polar transformation biologically plausible, or the useRetinalLogSampling should be used when we want plausible models?

Thank you,

[ctrl-z-9000-times]

[Retina Log Polar Sampling] as shown at:
http://users.isr.ist.utl.pt/~alex/Projects/TemplateTracking/logpolar.htm

The bottom of this webpage shows three images:
Cartesian image ----- Log-Polar representation ----- Retinal representation

• The Log Polar representation is biologically true.
• The "Retinal representation" is a reverse transform from the log-polar back to Cartesian coordinates. The purpose of showing the "retinal representation" is to visually demonstrate how/where the image is compressed.

[breznak]

[Retinal representation] The purpose of showing the "retinal representation" is to visually demonstrate how/where the image is compressed

ok, unless I'm doing something wrong, the images from retina (ie retina.getMagno())
are different when doing manually the log-polar transformation, and when using the useRetinalLogSampling.

The bottom of this webpage shows three images:
Cartesian image ----- Log-Polar representation ----- Retinal representation

it looks exactly like on this illustration, so it seems getMagno() is doing the "reverse transform from the log-polar back to Cartesian".

Do I need to use getMagnoRAW() not to do the reverse transform, in order to still

By using this [log-polar] transformation, both rotations and scaling between two images can be represented by translations.

?