Gaussian mask issue

Support
1

Hi Chris,

The attached video shows a mask issue we are seeing with one monitor.
This is an ASUS TN (twisted nematic) monitor, whose underlying tech is I believe 7-bit.
We are using std_dev 0.3 and mean 0.1 to generate this video (which gives a flatter-topped Gaussian mask).

It looks to me like there is a bit-depth or normalization issue in the mask that gives a large circle of constant alpha in the middle of the mask. Do you think that’s right? Is this something we can fix?

thanks,
Mark

2

Hi Mark,

I believe you’re seeing a consequence of the normalization term that I mentioned previously. For your given std_dev, the central region of the mask has values greater than one (see attached plot). However, the mask value is used as the output color’s alpha, which is clamped to the range [0,1]. Hence, every pixel in the central region gets an alpha of 1.

It seems like adding an option to disable the normalization term (as I’ve already proposed) would solve the problem.

Chris

3

Attachment: mask_value.png (54.8 KB)

4

Yes, such an option would be useful. Just to make sure we’re on the same page, the option I’d think should ensure the gaussian mask has alpha 1 at its peak.

I see you committed a change and I’m looking forward to testing.

Thanks,
Mark

5

Just to make sure we’re on the same page, the option I’d think should ensure the gaussian mask has alpha 1 at its peak.

Yes, exactly.

I see you committed a change and I’m looking forward to testing.

Yep. In the current nightly build, the drifting grating has a new parameter: normalized. To preserve the behavior of existing experiments, it defaults to YES. If set to NO (or 0, or false), the normalization term is omitted.

I’ve attached an example experiment that demonstrates the parameter’s effect, along with an image showing the stimulus display output (normalized Gaussian on the left, non-normalized on the right).

Chris

6

Attachments:

7

Thanks!
(MWEL, nice.)

Mark