Visual quality of Label about cesium

[pjcozzi]

If it fixes the problem, we should make it an option on the LabelCollection. This requires a minor shader tweak. Do you want to try it? I can help you.

[mramato]

So I took a crack at it and it definitely makes text look way better. The "quick" test I did was change line 46 in BillboardCollectionVS.glsl from

gl_Position = agi_viewportOrthographic * vec4(positionWC.xy, -positionWC.z, 1.0);

to

gl_Position = agi_viewportOrthographic * vec4(floor(positionWC.xy), -positionWC.z, 1.0);

Which now causes billboards to align to pixel. I assume the next step would be to add a set/get AlignToPixel method to Billboard and have billboard collection pass that to the shader. This would default to false, but LabelCollection would default it to true (since it uses a BillboardCollection underneath).

My main problem is that I have no idea what I'm doing. My first guess would be to turn originAndShow into originAndShowAndAlign and add a branching condition to call floor or not, based on the value of Alight, but I know branching conditions can be bad. Am I anywhere near close with my idea? How should I do this?

[pjcozzi]

As long as the branch is coherent, I wouldn't worry about it. Coherence is
easy to define for fragment shaders in screen space, but is less obvious
for vertex shaders; however, I assume it is simply the order of vertices in
the draw call. For kicks, alignToPixel could be zero or one, and we could
do lerp(positionWC.xy, floor(positionWC.xy) , alignToPixel) or other
tricks, but this isn't any better.

Can we just always align to a pixel? In what cases does this hurt visual
quality?

On Thu, May 10, 2012 at 6:48 PM, Matthew Amato <
[email protected]

wrote:

So I took a crack at it and it definitely makes text look way better. The
"quick" test I did was change line 46 in BillboardCollectionVS.glsl from

gl_Position = agi_viewportOrthographic * vec4(positionWC.xy,
-positionWC.z, 1.0);

to

gl_Position = agi_viewportOrthographic * vec4(floor(positionWC.xy),
-positionWC.z, 1.0);

Which now causes billboards to align to pixel. I assume the next step
would be to add a set/get AlignToPixel method to Billboard and have
billboard collection pass that to the shader. This would default to false,
but LabelCollection would default it to true (since it uses a
BillboardCollection underneath).

My main problem is that I have no idea what I'm doing. My first guess
would be to turn originAndShow into originAndShowAndAlign and add a
branching condition to call floor or not, based on the value of Alight, but
I know branching conditions can be bad. Am I anywhere near close with my
idea? How should I do this?

---

Reply to this email directly or view it on GitHub:

#15 (comment)

www.seas.upenn.edu/~pcozzi/

[mramato]

Aligning to pixel can make moving billboards appear jumpy as they snap from pixel to pixel, I notice this on my machine. We may be able to get the "best of both worlds" by only rounding the y component. This will still make the text sharper, but also cut down on the jumpiness. I tried this out on my machine and I think it might be the best solution.

gl_Position = agi_viewportOrthographic * vec4(positionWC.x, floor(positionWC.y), -positionWC.z, 1.0);

[pjcozzi]

Sounds like a good idea. We can always add an option later. If we add the
option, we can consider doing at the collection level, not per billboard.
This way we would not need to store the flag per vertex; instead, it would
be a uniform.

On Fri, May 11, 2012 at 11:40 AM, Matthew Amato <
[email protected]

wrote:

Aligning to pixel can make moving billboards appear jumpy as they snap
from pixel to pixel, I notice this on my machine. We may be able to get
the "best of both worlds" by only rounding the y component. This will
still make the text sharper, but also cut down on the jumpiness. I tried
this out on my machine and I think it might be the best solution.

gl_Position = agi_viewportOrthographic * vec4(positionWC.x,
floor(positionWC.y), -positionWC.z, 1.0);

---

Reply to this email directly or view it on GitHub:

#15 (comment)

www.seas.upenn.edu/~pcozzi/

[mramato]

Cozzi, assuming there are no issues with my change, can you close this out? Thanks.

[pjcozzi]

Looks good. Good catch updating the JavaScript code that is used to compute the screen-space position of a billboard.