Fog

[bagnell]

• Tweak fade distance and density
• Change tile SSE when in fog
• Use preprocessor instead of conditional when enabled.
• Swap shader when a tile is or is not in fog
• Performance
• Tests
• Reference doc including advice on the performance / visual quality trade-off
• Update CHANGES.md

[pjcozzi]

Consider using the same views we used for OBBs: http://cesiumjs.org/2015/06/24/Oriented-Bounding-Boxes/

Instead of commenting them out, they could be in a drop down for future testing.

[pjcozzi]

Added a few items to the tasklist

• Tests
• Reference doc including advice on the performance / visual quality trade-off
• Update CHANGES.md

[pjcozzi]

@bagnell add any ideas we come up with that are outside the scope of this PR to the renderer roadmap, #3001.

[pjcozzi]

Let's move the fog equation to a @private helper function. Keeping it in this file is probably fine for now.

[pjcozzi]

Zoom in the part of the globe that is dark due to the ground atmosphere. Note how the fog blacks out the distant terrain.

With fog:

Without fog:

[pjcozzi]

How hard is it to use the ground atmosphere if the ray hits the ellipsoid, but the sky atmosphere if the ray doesn't?

[pjcozzi]

What are your stats for the default view in the fog Sandcastle example.

With ~3K resolution:

No fog:

Visited 751, Rendered: 498, Culled: 263, Max Depth: 20, Waiting for children: 0

Fog:

Visited 719, Rendered: 477, Culled: 251, Max Depth: 20, Waiting for children: 0

Not great.

[pjcozzi]

For the night ground atmosphere, we might want to use the negative sun vector or something. For example:

No fog:

Fog:

Note that for this view, fog is pretty useful.

No fog:

Visited 140, Rendered: 86, Culled: 78, Max Depth: 10, Waiting for children: 0

Fog:

Visited 100, Rendered: 63, Culled: 50, Max Depth: 10, Waiting for children: 0

[pjcozzi]

Overall, I am not impressed with the number of tile reduced. I expected a lot more. Perhaps we can increase the start of the step function and decrease the completely-in-fog distance.

I also imagine the VMSSE-style SSE based on fog will be somewhat effective; however, it's effectiveness will depend on how many tiles are partially in fog. Above, I am suggesting to put more tiles completely in fog, but fewer partially in fog...

[mramato]

When this is done, it would also be interesting to see numbers for when maximumScreenSpaceError is set to 1 instead of it's current default of 2, since 1 is the value we want to be able to use for the default (but need to bring the number of tile requests down in order to do it).

As an aside, is the minimum value for maximumScreenSpaceError 1 or 0? I always assumed it was 1 but the doc doesn't specify.

[kring]

As an aside, is the minimum value for maximumScreenSpaceError 1 or 0?

Any value > 0 is allowed. 0 makes no sense, but less than 1 is fine. It doesn't have to be an integer.

You can also think of it as a range. If your maximumScreenSpaceError is 2, your texels will be between 1 and 2 pixels on the screen. If maximumScreenSpaceError is 1, they'll be between 0.5 and 1 pixels on the screen. 1.5 might be a nice value for the default, giving you between 0.75 and 1.5 pixels.

[mramato]

Any value > 0 is allowed. 0 makes no sense, but less than 1 is fine. It doesn't have to be an integer.

I'll write up an issue to clarify the doc. We might also want to consider adding a get/set property to enforce non-zero values.

[pjcozzi]

The tile stats for various horizon views are awesome - often 33-50% few tiles without much visual quality impact.

[pjcozzi]

Default density and SSE factor seem sensible. Increasing the SSE factor can result in significant savings (like another 33%), but, of course, hurts visual quality and makes popping more obvious when the camera moves.

[pjcozzi]

@bagnell can you do a quick sanity check for when the camera is really low or has a negative altitude?

For a horizon view in the dead see, without fog:

Visited 236, Rendered: 148, Culled: 118, Max Depth: 14, Waiting for children: 0

With fog:

Visited 47, Rendered: 21, Culled: 59, Max Depth: 14, Waiting for children: 0

That seems too aggressive.

[pjcozzi]

For a horizon view at a high altitude, I am seeing a lot of cracking between the globe and the sky atmosphere. It loads eventually, but I wonder if we are too aggressive at this altitude.

[pjcozzi]

Wow, check out these stats for a horizon view near the ground in FL:

No fog:

Visited 352, Rendered: 226, Culled: 154, Max Depth: 18, Waiting for children: 0

Fog:

Visited 139, Rendered: 82, Culled: 91, Max Depth: 18, Waiting for children: 0

No difference in visual quality.

[pjcozzi]

For a horizon view at a high altitude, I am seeing a lot of cracking between the globe and the sky atmosphere. It loads eventually, but I wonder if we are too aggressive at this altitude.

Also follow the ISS or GeoEye satellite in the CZML Sandcastle example to see this a bit.

[pjcozzi]

How many tiles are saved for a horizon view like this in space? Perhaps we should just turn it off at this altitude.

[bagnell]

@pjcozzi I addressed all of your comments. This is ready for another look.

[pjcozzi]

I get this exception when I open Hello World in Edge in this branch, but not master. Perhaps the other browsers are not checking the GLSL as carefully or is this a bug in Edge?

[pjcozzi]

Actually, Firefox also doesn't like it:

[pjcozzi]

2D is fixed. Looks like fog is also disabled in Columbus view. Does it need to be? It appeared to work when I tested it last.

[pjcozzi]

Since we removed the discard statements for the atmosphere, the atmosphere can look a bit rough when waiting for the initial globe tiles to load:

We could just leave this or not draw the atmosphere until the globe renders something. I can't come up with a reason that drawing the atmosphere without the globe is useful.

[pjcozzi]

Also, for the blog post, can you include FPS (if you can get useful data) in addition to the number of tiles?

[pjcozzi]

Just those comments and this is ready!

[bagnell]

@pjcozzi This is ready.

[mramato]

1. Did we test this on mobile?
2. Should we add a toggle fog button to the Terrain example? I think it would come in handy.
3. Some views still look really washed out, take this shot with and without fog near San Francisco Bay. Maybe this is just the nature of fog? But it makes some views pretty bad.

Without

With

[pjcozzi]

Tests look good now.

Did we test this on mobile?

Good idea.

Should we add a toggle fog button to the Terrain example? I think it would come in handy.

Also, a good idea.

Some views still look really washed out, take this shot with and without fog near San Francisco Bay. Maybe this is just the nature of fog? But it makes some views pretty bad.

This is OK with me. The less aggressive we make it, the less performance will improve. This was already tweaked quite a bit, and is a reasonable default. We can change it later if needed.

Let me know when this is ready.

[bagnell]

@pjcozzi This is ready.