Jump to content

View more

Image of the Day

#ld38 #screenshotsaturday Mimosa Fizz action gif #2 https://t.co/TUzdppvfUL
IOTD | Top Screenshots

The latest, straight to your Inbox.

Subscribe to GameDev.net Direct to receive the latest updates and exclusive content.

Sign up now

Changing LoD on instancing and alpha testing

4: Adsense

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
2 replies to this topic

#1 QNAN   Members   


Posted 30 June 2013 - 08:08 PM

Im working on instancing and would like to allow the instancing area (the area holding all instances) to change LoD, if the player moves closer/further away.


Since alpha blending is a no-go with instancing as it is impossible to sort render order with them, I have experimented with alpha testing and dissolvemap.


My problem so far is, that the dissolving matches up pretty bad across the fading time. Last test I tried with this calculation on the shader:

l_final_color.a = l_original_alpha * saturate(g_alpha * (0.5f + tex2D(g_dissolvemap_sampler, a_texcoord0).r))

g_alpha is the fading over time (will be 0.1 at 10% of the time etc). The dissolvemap, which is a noisemap with values distributed across the 0-1 spectrum. Alpha test will succeed when value is above 0.5.

I offset it with 0.5 to get the median values to fade in at around 0.5. If I do not use this offset, the first values (those that are 1.0 on the dissolvemap) will first start trickling in at 0.5. By offsetting it, I can get them to fade in at 0.333~. Which is also quite late...


To close the 0.333~ gap, I offset the g_alpha by 0.333~ from the main program, while scaling the timing by 0.666~, effectively removing the start gap while keeping the timing intact.


Anyway, the fading kinda works, but the bulk of the fading takes place in a too short time span. I have watched it up-close and it does fade over the entire time, but the beginning kinda trickles in and is not noticable from afar, and similar with the ending, making both beginning and ending not noticable. The result is a fast fade, that looks too much like a pop.


So my questions are:

- Is it possible to make a better distribution equation for the final color? (I have tried with exponentials but couldn't arrive at anything satisfactory)

- Can the problem be the dissolvemap? (IMO it looks fairly evenly distributed, I attached it (original is in .bmp, I saved as .jpg to save space here)


Is there alternatives to this approach?

Attached Thumbnails

  • dissolve_map.jpg

#2 MJP   Moderators   


Posted 01 July 2013 - 01:51 AM

What you basically want is the alpha to be a probability that a pixel is opaque. I would do it like this:

float random = tex2D(g_dissolvemap_sampler, a_texcoord0).r;
if(alpha < random)

You can also use alpha-to-coverage, which will basically accomplish the same thing using a tiled screen-space dither pattern instead of a pure random pattern. You can also  encode a set of dither patterns directly into a texture, and then lookup into that texture based on screen position and alpha value.

Edited by MJP, 01 July 2013 - 03:58 PM.

#3 QNAN   Members   


Posted 01 July 2013 - 09:48 AM


Thanks, it was something like this I needed smile.png

Your code makes sure, that values are included right away. Can't believe what a mess I ventured into with the other code (which I took over from a far-range fadeout of instancing with too little tthought).


Just for the record, the final shader code will look like this:

float l_dissolvemap_result = tex2D(g_dissolvemap_sampler, a_texcoord0).r;
if(g_alpha < l_dissolvemap_result)

l_final_color.a = l_original_alpha;

In this way alpha-testing (or alpha-to-coverage, which is very similar, just with nicer edges and antialiasing) will still work for the original alpha of the object texture, as the g_alpha is not factored in.

Edited by QNAN, 01 July 2013 - 11:00 AM.

Old topic!

Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.