Deep camera maps

On this page

Overview
How to
DCM tips
Using deep maps in shaders

Overview

Deep camera maps are rendered images, where semi-transparent areas (such as volumes) between the camera and the nearest opaque surface are stored with depth information. Each pixel in the image is represented as a curve describing how the transparency value changes across the depth of the scene. This allows you to composite rendered images and have the semi-transparent areas blend correctly according to their depth.

Deep camera maps are useful for objects with high transparency across the object, such as swarms of sprites and clouds (volumes). You should render each such object as a separate DCM. DCMs let you render transparent objects separately from the objects that interact with them, allowing much more flexibility in compositing.

The main drawback of deep camera maps is high disk space overhead, given the amount of information stored per pixel. A 2K (2048×1080) DCM frame can typically take 1-2 GB of disk space, and the typical space budget for a shot might be 1 TB. Another drawback is that, since they are added in compositing, objects in DCMs cannot receive shadows.

Houdini only supports storing DCMs in the .rat or .exr texture file format. The same image can contain a deep camera map (depth per pixel information) and deep rasters (extra image planes).

How to

To render a deep camera map, you must add two rendering properties to the camera or render driver node.

To...	Do this
Set up a camera or render output node to render as a DCM	Select the camera or render driver node. In the parameter editor, click the Gear menu and choose Edit rendering parameters. In the tree sub-pane on the left, open the "Mantra" folder, then "Output". Select the DCM Filename and Deep Resolver properties and click the button to move them to the Existing parameters side. This adds the properties to the node’s parameter interface. Click Accept. In the node’s parameter editor, click the Output tab. (On a render driver node, Output is a sub-tab of Properties.) Set the Deep resolver parameter to Deep shadow resolver. Set the DSM filename to the name of the RAT or EXR file you want to save the into, for example `$HIP/textures/$OS_$F_dcm.rat`. (`$HIP` is the directory containing the current file. `$OS` is the name of the current node. `$F` is the frame number.) When you render, mantra will write out the DCM in addition to the output image.
Customize trade-offs between accuracy and file size	Several properties allow you to decrease depth compositing accuracy in order to reduce the size of DCM files. If the object in the DCM is not detailed or the distance between the near and far clipping planes is small, depth compositing accuracy is less important and these properties can be very useful. Add the DCM compression property and increase it to throw out more samples. Add the DCM Z Storage property and change it from "32 bit float" to "16 bit float". This halves the number of bits used to represent the depth of samples. If this setting is too low, you will see stair-stepping and "temporal buzzing" in animation. Add the DCM Z-bias property and increase it. Samples within this value of each other are merged.
Composite two rendered DCMs together	Use the DSM Merge render node or the dsmmerge command line utility.
Convert a DCM file into a flat image	Use the DSM Flatten compositing node.

DCM tips

Pre-multiply the diffuse component by the opacity in the shader.
Every object needs its own DCM.
Use $OS (name of the current node) in the DCM file name to keep DCMs separate.

Using deep maps in shaders

In VOPs, see the Shadow Map VOP,

In VEX, see teximport, shadowmap, and dsmpixel. For example:

cvex 
dsminfo(string map="foo.rat") 
{ 
   string   names[], name; 
   float   Pz[]; 
   vector   Of[]; 
   vector   res; 
   int      x, y; 

   if (teximport(map, "texture:resolution", res)) 
       printf("%s resolution: %g\n", map, res); 
   teximport(map, "texture:channelnames", names); 
   printf("%d channels:\n", arraylength(names)); 
   foreach (name; names) 
       printf("\t'%s'\n", name); 

   x = res[0] * .5; 
   y = res[1] * .5; 
   if (dsmpixel(map, "Pz", x, y, Pz) && 
       dsmpixel(map, "Of", x, y, Of)) 
   { 
       int      i, n; 
       n = arraylength(Pz); 
       printf("Depth complexity[%d,%d] = %d\n", x, y, n); 
       for (i = 0; i < n; i++) 
      printf("\tZ=%g, Of=%g\n", Pz[i], Of[i]); 
   } 
}

Deep camera maps

Overview

How to

DCM tips

Using deep maps in shaders

Rendering

Mantra user guide

Basics

Lighting

Next steps

Guru-level

Other renderers