Gabor (Grating) size and colors in OpenSesame 4

Wouter

Hey all,

I'm running OpenSesame 4 and I've been having some trouble getting circular grating stimuli with exactly the size I want. Running the following in psychopy gives me a grating stimulus that is just a tiny bit bigger than the circle. The difference seems to scale with stimulus size, and I'm not sure where this factor could be coming from.

# this is in my simple, single inline script item:
x  = Canvas()


x.gabor(0,0, size=60, env='circular',orient=10,freq=1)
x.circle(0, 0, 30,color='darkred')
x.show()

kb=Keyboard()
kb.get_key()

This is happening with psychopy; I dove all the way to the psychopy source, but was not able to find out where the additional size came from.

I tried to test this with different backends, but then I'm running into a different problem. In both legacy & Expyriment, the code crashes on the fact that whatever color I'm providing for col1, it states it's not a valid color format (white, #ffffff, 255, or tuples). I don't have this problem for other stimuli that I'm trying to give a color; x.fixdot(color='white') runs perfectly fine.

Any help is appreciated!

Wouter

Wouter

I've done some digging in the psychopy documentation, and my eye landed on a sentence that the underlying texture of a gratingStim should be a square 'power-of-two' image; and in the source I saw comments that pointed to this as well.

I started to suspect that maybe psychopy scales the texture first, and then draws the size of the mask as a function of the size of the texture; For gabors this isn't a problem, but for circular gratings it means that the size is dependent on the nearest power of two.

An updated version of my script shows that this indeed seems to be happening; it draws gratings at increasing sizes, and draws a red circle at the expected size on top of it.

# this is in my simple, single inline script item:
import numpy as np
x  = Canvas()

sizes = np.arange(50, 512) ## watch the size of the gabor increase in jumps.
# sizes = 2**np.array([4,5,6,7,8,9,10]) ## for 2**[even numbers] it actually seems to match
for gsz in sizes :
    x.gabor(0,0, size=gsz, env='circular',
            orient=10, freq=0.05)
    x.circle(0, 0, gsz//2,color='darkred',penwidth=3)
    x.show()
    
    print (gsz)
    kb=Keyboard()
    # cycle faster or slower, dependent on the number of size-steps
    kb.get_key(timeout=(7500//sizes.size))
kb.get_key()

test.osexp

As for the color issue: I have no idea why, but it seems like the error is raised in opensesame's Legacy color backend, line 99. That suggests that for whatever reason neither 'white' nor '#ffffff' are passed down as strings.

Wouter

About psychopy and sizes of circular gratings

I did some digging and got the suspicion that psychopy scales the size of the texture that is underlying a grating up to the nearest power of 2, and then determines the size of the circular mask based on the size of the texture. I couldn't find exactly where this happens, but I saw comments like this in the source

I tested it with the following script, and there seems to be some truth to this... the size of the red circle here grows smoothly, whereas the size of the grating changes in increasingly big jumps:

# this is in my simple, single inline script item:
import numpy as np
x  = Canvas()

sizes = np.arange(50, 512) ## watch the size of the gabor increase in jumps.
# sizes = 2**np.array([4,5,6,7,8,9,10]) ## for 2**[even numbers] it actually seems to match
for gsz in sizes :
    x.gabor(0,0, size=gsz, env='circular',
            orient=10, freq=0.05)
    x.circle(0, 0, gsz//2,color='darkred',penwidth=3)
    x.show()
    
    print (gsz)
    kb=Keyboard()
    # cycle faster or slower, dependent on the number of size-steps
    kb.get_key(timeout=(5500//sizes.size))
kb.get_key()

About colors in legacy/xpyriment:

I can't understand why, but the error is triggered in the legacy color module, line 99 . Somehow this means that my color strings are at some point no longer strings..

eduard

Hi Wouter,

Can you hack the source code to find out what your color strings changed to before the exception is thrown?

Also, it sounds as if these problems are specific to OS 4. Is this the case, or do they also occur in OS3? I'd be surprised if the weird gabor behavior occurred only in OS4.

Eduard

Wouter

Hi Edu! Thanks for your response. I'm indeed not entirely sure these things are OS4 specific. But I suspect they have to be 'recent'.

Size in psychopy

I suspect this isn't OS4-specific, but I'm surprised no one seems to have brought it up before. Maybe it's something that was introduced with psychopy 3 ?

If I'm correct about what's happening here, I was going to propose that we could create a custom numpy mask for circular gratings, and pass that to psychopy rather than the label 'circ'. In general, the circular mask created by default seems to scale poorly (very jagged edges with larger gratings). That could then be improved in a single go. Cons are that there's something ugly about making 'circular' gratings an exception, and it'll probably affect the speed at which they are drawn.

Within the experiment we're building, we're now hacking around this by drawing a fat open circle around the grating in the color of the background. It's ugly, but it works.

An alternative would be to bother Jon Pierce about this to get a better understanding (and maybe ask them to clarify the documentation).

But before I'd start developing a solution, I wanted to post it here first and see whether there's something weird about my installation or about my thinking.

Gabor/Noisepatch Colors in Legacy & Expyriment

print(type(colorspec)) returns a <class 'openexp._color.legacy.Legacy'>. colorspec is the second argument in the constructor. I suspect something is going wrong in the steps leading up to it, when passing around *args.

I tried to pinpoint it, but I get very lost in the modern way (👴) that canvas elements are drawn. I suspect @sebastiaan would know exactly where to look.

eduard

An alternative would be to bother Jon Pierce about this to get a better understanding (and maybe ask them to clarify the documentation).

+1

Though, I guess, the power of 2 setup of Psychopy is something they wouldn't want to change all too lightly. But all that can be done here is hacky in one way or another, no?

sebastiaan

Hi @Wouter and @eduard ,

Thanks for spotting these issues. Both the color-conversion issue with the legacy and xpyriment backends and the size issue with the psycho backend are related to OpenSesame, not PsychoPy. I will publish a bugfix release (4.0.13) to address them later today!

— Sebastiaan