yes, the way he does this is quite weired. unless it's for some sort of special effect.
but ideally, you only want overbright pixels (> 1 after exposure multiplication) to produce a glow, so you should clamp to 1 and substract 1 (or the other way round, then blur and add it to the frame buffer)

ah, yes, a lerp could explain it, although I don't really see the benefit, it just seems overkill

lerp is slang for linear interpolation. Basically, it's a function that lets you transition smoothly from color a to b. So, it you lerped between a pixel with color 0,0,0 and a pixel 0,0,200, the pixel halfway between them would be 0,0,100.

The final image has to be darker than the original, so we can display it. As he says, the step where lerp is used is the tone mapping step, which is used to make the map displayable.

Edit - with regards to your first question. Maybe I'm getting confused from your use of the word blend. But in this particular example, the only suppresion of LDR values is when creating the image to be blurred. So he takes the original, removes LDR values, then blurs it. Then lerps the original values with the blurred values, and since LDR values were removed from the image that was blurred, we have very little if any blur around the LDR values in the resulting image, which stands to reason.

Quote:Original post by mikeman

Quote:Original post by digitec devil
The final image has to be darker than the original, so we can display it. As he says, the step where lerp is used is the tone mapping step, which is used to make the map displayable.

I just gave an example when the original color of the pixel whould be (0.8,0.8,0.8).We can display that.But,the way he supresses LDR,it will be lerped with (0,0,0),thus darkening it for no reason.

The middle step you're leaving out is the blurring step, while in some cases you're right it may be lerped with (0, 0, 0), we can't guarantee that.

ah, I just had a look at the code of that tutorial.

I didn't understand you when you said this:
"If we had (0.8,0.8,0.8),the program would dismiss it,but if we had(1.2,0.8,0.8),it would keep the whole color,not just the red channel.I thought the proper way to do it was to subtract 1 from all components and clamp to 0."

if you're wondering why he does this:

float4 SuppressLDR( float4 c )
{
if( c.r > 1.0f || c.g > 1.0f || c.b > 1.0f )
return c;
else
return float4( 0.0f, 0.0f, 0.0f, 0.0f );
}


it's just to keep a correct glow colour. you can't consider each pixel's channels separately (if that's what you wanted to do), it will produce incorrect colors.
if you've got a color shade that goes from 0.9, 0.5, 1.6 to 2.3, 1.2, 3.1 (r ,g ,b)

considering each color separately will produce color bands in the glow, as each channel will go above 1.0 separately, not at the same time. the first glow color you'll get for low intensities will be blueish, then when the red overflows it will look purple-ish, then after the green component overflows you'll have to add it to the glow color too, and each time a component gets above 1, you will get color banding. (not really as noticeable as regular color banding of course, as there is a blur applied to it, but still)

EDIT: formatting