Selection adds semi transparent pixels when entering Refine but making no changes

[thomaso]

I tried to reproduce the difference in selection discussed in this thread … and run into an unexpected result:
Below the green layer got created from a flood-selection of the blue + fill colour. – But what happens now when I flood-select the green …?

Edited September 26, 2022 by thomaso
+ link to initial thread

[NotMyFault]

4 hours ago, thomaso said:

Below the green layer got created from a flood-selection of the blue + fill colour. – But what happens now when I flood-select the green …?

Could you upload the actual file, or the screen recording of the steps to create the green layer?

It looks like the green pixel layer has a small halo, maybe the selection it was based  on had feathered edges. In the file you could create a spare channel of the selection, create a mask layer, and the inspect for actual alpha values to spot the cause.

Take into account that the marching ants gives absolutely no indication about hard / soft edges of a selection. It is merely the line between below and above 50% selected / alpha.

[Dan C]

@thomaso does your document have bleed enabled? As I can see you are using StudioLink in Publisher and I believe there is a known bug logged with the development team regarding offset selections with bleed enabled.

I suspect the fact one layer is an (Image) and the other is a (Pixel) would explain why one exhibits the pixel shift, and one does not

[thomaso]

6 hours ago, NotMyFault said:

Could you upload the actual file, or the screen recording of the steps to create the green layer?

2 hours ago, Dan C said:

bleed enabled?

It was created with a pixel selection of the blue area > new pixel layer > fill. I can't recreate an object like this. In a trial before recording the video below hiding all channels the Alpha excepted changed this selection to normal (green only) and could not get undone (!) when all channels got activated again, which might indicate a rendering / redraw issue. But in the clip this procedure did have such a 'healing' effect and the issue still occurs.

There was no bleed set, neither in the initial .afpub when this object was created nor in the new document I copy/pasted it.

pixel selection issue.afpub

[NotMyFault]

If you open the file with Photo / channels panel, you can spot spray pixels in green channel outside, and spray alpha values (below 255) inside of the head.

i typically get such effect when using „refine“ on a selection, but the patters is a bit different as the spray density seems unrelated to distance.

Maybe an HW accel issue or cosmic radiation corrupting memory.

 

BTW: Photo on iPad crashed a few seconds after opening the file, but worked in 2nd attempt.

 

[thomaso]

21 minutes ago, NotMyFault said:

i typically get such effect when using „refine“ on a selection, but the patters is a bit different as the spray density seems unrelated to distance.

Very good, that's it! Your hint reminded me to have experimented with Refine last night to detect a possible reason for the difference in the OP's two screenshots PSD vs. AfPh. – This appears to make it reproducible now.

Whereas the resulting obvious offset to top left feels rather like a bug, (and I wonder about tiny deviations at certain spots of the selection ants compared to the coloured pixels), @Dan C.

[Dan C]

Thanks for your file and screen recording providing Thomaso, I've found a few oddities here:

• In a new document, following your exact workflow on mac or Windows, I'm unable to replicate the behaviour.
• Using your file, when trying to flood select on the Image layer (which is a missing resource), the app crashes. I can replicate this in a new file with a different missing resource, therefore I've logged this crash with the devs as a bug.
• Flood selecting the Pixel layer you created from your selection, I see the same offset selection shown above, but again I'm unable to replicate this in a new file.

Unfortunately due to the above crash, I can't follow your workflow in the document you've provided - can you please try this in a new file, with a new screenshot and let me know if this still occurs?

If so, could you provide a copy of this new document with the Image layer embedded?

[NotMyFault]

4 hours ago, thomaso said:

Whereas the resulting obvious offset to top left feels rather like a bug, (and I wonder about tiny deviations at certain spots of the selection ants compared to the coloured pixels),

There is no offset. As i explained before, there are a lot of stray pixels, actually not symmetric so it looks similar to an offset. Such asymmetric / random stray pixels are a common observation when using refine.

what is uncommon: there are green pixels (right side) despite alpha channel is 0. Normally Affinity immediately deletes RGB values when alpha is 0.

[thomaso]

2 hours ago, Dan C said:

Unfortunately due to the above crash, I can't follow your workflow in the document you've provided - can you please try this in a new file, with a new screenshot and let me know if this still occurs?

The blue screenshot was just downloaded from this OP's post. – I can reproduce the selection oddity with a different screenshot, while it behaves different in different .afpubs. In the clip below again: First in my permanent test file, then in the recently created + attached / updated new document.
Not in the clip: I copied the finally resulting pixel layer from the 1st .afpub to this 2nd .afpub for your closer comparison of both results.

pixel selection issue.afpub

 

34 minutes ago, NotMyFault said:

There is no offset. As i explained before, there are a lot of stray pixels, actually not symmetric so it looks similar to an offset.

Maybe, to me it still seems to be but it might change with more experiments. While the first experience (blue cow image) looked quite obvious like an offset to top left (though the image is symmetrical) the new tests first had a tendency to left again (where coloured pixels are) but then rather equal around the circular shape. – Did you experience this increased selection also to right or bottom direction yet?

[NotMyFault]

The issue with stray pixels / halo after flood selection / refine / new pixel / fill occurs when the source color has some noise. It does not occur if the source is solid pure color, or not refined.

1. use Photo
2. Create circle (or other shape)
3. fill in blue
4. rasterize
5. add noise 6%
6. rasterize
7. flood select 
8. do not refine
9. add pixel layer
10. flood fill in green
11. use channels panel, select green only -> all well / hard edges
12. deselect
13. delete green layer
14. activate blue layer
15. flood select
16. now refine
17. add pixel layer
18. fill in red
19. activate red channel
20. see many red stray pixel. Location / distribution depends on shape.

video has some mis-clicks in the middle, but shows the difference between red (refined) and green (no refine).

[Dan C]

Thanks Thomaso for the further info & docs provided - I can still see the issue with the layers you created in your file, but following the exact workflow on both Windows and Mac doesn't produce the same issue for me in your file, the filled Pixel layer created is exactly the shape of either selection with no additional semi-transparent pixels for me.

Can you please provide a screenshot of your Performance Preferences, so I can ensure I have exactly the same setup here?

[thomaso]

1 hour ago, Dan C said:

the filled Pixel layer created is exactly the shape of either selection with no additional semi-transparent pixels for me.

Can you please provide a screenshot of your Performance Preferences, so I can ensure I have exactly the same setup here?

Sure you did Refine (with its default values 10, 0, 0, 0, 50)? Its quickmask + Apply don't not show the later ant oddities, only a new flood select after fill. Then the coloured pixel are visually more obvious if I hide the screenshot.

Regardless of this ant oddity, I wonder why it behaves differently in 2 different documents. So I just did another test on the top left corner of my 2-page .afpub to see if the size of the spread might have an effect. It doesn't. But today, the 1st of 3 attempts at 2 spread positions showed Refine's quick mask as a fuzzy halo around the circle and its inner part - while the 2nd and 3rd attempts showed it as a sharp mask, just like every attempt yesterday.

[Dan C]

1 hour ago, thomaso said:

Sure you did Refine (with its default values 10, 0, 0, 0, 50)? Its quickmask + Apply don't not show the later ant oddities, only a new flood select after fill. Then the coloured pixel are visually more obvious if I hide the screenshot.

Absolutely, however I've since reset Publisher using the CTRL run method and I'm now able to replicate this on macOS - I'm not 100% sure why I wasn't seeing this previously, but now I'm able to replicate it I can get it logged with our devs

1 hour ago, thomaso said:

But today, the 1st of 3 attempts at 2 spread positions showed Refine's quick mask as a fuzzy halo around the circle and its inner part - while the 2nd and 3rd attempts showed it as a sharp mask, just like every attempt yesterday.

That's certainly strange, I haven't seen that during any of my testing but should it appear again I'd recommend creating a new bug report for this!

[thomaso]

3 hours ago, Dan C said:

it logged with our devs

Dan, when I asked my initial question in this thread I did not expect it would cause so many posts – which actually aren't related to the OP's question. – Would you mind to split this thread and move my first post + the part below to the bugs forum?

[Dan C]

@thomaso of course, this is now a new thread in macOS Photo Bugs - you will need to follow the thread if you wish to be notified of further replies