Jump to content

  • Log In with Google      Sign In   
  • Create Account


HDR Bloom with RGBM encoding format


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
3 replies to this topic

#1 jeremie009   Members   -  Reputation: 255

Like
0Likes
Like

Posted 21 September 2012 - 11:31 PM

I've been trying to achieve bloom using a 8bit render target using rgbm but after I failed several times I've been wondering if it was even possible.

If i'm not mistaken the alpha contain the data about the intensity so when I doing the offset to do my Gaussian blur am I messing up the encode information ?

here's the code I'm using to encode and decode my image

float MaxRange = 6;
float3 DecodeRGBM(float4 rgbm)
{
return rgbm.rgb * (rgbm.a * MaxRange);
}

float4 EncodeRGBM(float3 rgb)
{
float maxRGB = max(rgb.x,max(rgb.g,rgb.b));
float M = maxRGB / MaxRange;
M = ceil(M * 255.0) / 255.0;
return float4(rgb / (M * MaxRange), M);
}


I'm able to encode and decode any image but when I try to blur the image doesn't look right. Highly saturated and lot of crazy color.


And here's my blur function
( taken from matt pettineo)

/ Performs a gaussian blue in one direction
float4 Blur(float2 texcoord, float2 texScale, float sigma)
{
float4 color = 0;
for (int i = -6; i < 6; i++)
{
float weight = CalcGaussianWeight(i, sigma);
float2 texCoord = texcoord;
texCoord += (i / size) * texScale;
float4 sample = tex2D(BloomMap, texCoord);
sample.rgb = DecodeRGBM(sample);
sample.a = 1;
sample*=sample;
color += sample * weight;
}
return color;
}

I also I'm using xna and I'm able to blur the image using hdrblendable but Id prefer to use an 8 bit image so I can use a filter on it. ( xna doesn't support filter anymore for many format).

Sponsor:

#2 Hodgman   Moderators   -  Reputation: 29671

Like
0Likes
Like

Posted 21 September 2012 - 11:40 PM

when I doing the offset to do my Gaussian blur am I messing up the encode information ?

You can't bilinearly filter or blend RGBM values.
e.g. 10*10 and 5*20 are the same, but if I blend between them I get 7.5*15 which is different.

So, yeah, if your bloom samples are in-between pixels, then you'll be doing strange things to your colour values. You could try making the BloomMap sampler use nearest/point filtering.

[edit]Also, at the end, instead of return color; you probably need return EncodeRGBM(color.rgb); so that the next blur pass works.

Edited by Hodgman, 21 September 2012 - 11:43 PM.


#3 mhagain   Crossbones+   -  Reputation: 7863

Like
0Likes
Like

Posted 22 September 2012 - 06:26 AM

Instead of storing a multiplier in the alpha channel, store a divisor; it will work fine with linear filtering (I haven't tested it with blending though).

It appears that the gentleman thought C++ was extremely difficult and he was overjoyed that the machine was absorbing it; he understood that good C++ is difficult but the best C++ is well-nigh unintelligible.


#4 jeremie009   Members   -  Reputation: 255

Like
0Likes
Like

Posted 22 September 2012 - 01:44 PM

@ Hodgman thanks for the answer. I'm encoding it back to rgbm but I forgot to put the function in my code.

@ mhagain. I'll try the divisor see if it works.

Since the rgbm is also storing some of the luminance info in the rgb that might be the real problem.

But still I'm wondering how the cryengine guy did their motion blur using rgbm format. I'm getting some pretty harsh line on my bloom when I use the point filtering.




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.



PARTNERS