Skip to content

Latest commit

 

History

History
55 lines (43 loc) · 3.59 KB

sdf-editor.md

File metadata and controls

55 lines (43 loc) · 3.59 KB

Index | Usage | SDF Editor

SDF Editor

Getting Started

  1. Build Falcor.
  2. Launch Mogwai.
  3. Load the script data/sdf-editor/SDFEditor.py. This uses the PathTracer, AccumulatePass, ToneMapper, and there is an SDFEditor pass at the end.
  4. Load the scene data/sdf-editor/SDFEditorStartScene.pyscene. You should see a single sphere in the center of the windows. Now you can start using the SDF editor. Pull down the GUI for the SDFEditor RenderPass. Hoover with the mouse over the tooltop, i.e., the "(?)" at the top to get instructions on how to use the editor.
    • ALT lets you add/subtract geometry while pressing the left mouse button.
    • TAB brings up the GUI for selecting which primitive and which primitive operation.

File formats

There are two types of SDF file formats that Falcor currently supports:

  • .sdf: That stores a list of 'edits' as a text file, and
    • Note that SDFEditorStartScene.pyscene (see Getting Started) loads the single_sphere.sdf, which contains just a single sphere.
    • You can change so that it loads test_primitives.sdf instead to see other primitives.
  • .sdfg: That stores the signed distance field as a binary file.

However, the SDF editor only supports loading the .sdf format, but can save as a .sdfg file (this is likely changing).

The SDF Editor RenderPass

The SDF Editor is implemented as a render pass, the inputs and outputs are as follows:

Name Type Description
vbuffer Input Encodes the primary hit points.
linearZ Input Linear depth.
inputColor Input Color texture from a previous render pass.
output Output Final color texture.

A simple render script with the SDF Editor using the path tracer can be seen in data/sdf-editor/SDFEditor.py.

To begin editing with the SDF editor, you need to have at least one SDF grid instance of the type SBS in the scene. The SDF grid is created by calling SDFGrid.createSBS(). It takes two optional arguments. One for the brick width (brickWidth), which is the number of voxels it should pack into one AABB to be ray traced, the other argument is if it should compress the data or not (compressed). The create function returns a SBS grid object that can be added to the scene by calling addSDFGrid() from the SceneBuilder.

If you do not want to start from scratch you can load a SDF file from memory by using the function loadPrimitivesFromFile from the sdf grid object. This takes in the path to the file (path) and the grid width (gridWidth). The grid width is the desired resolution of the grid (the internal implementation might increase the grid width if it does not align with the brick width).

Data structure creation in pyscene-files

The typical setup is:

sdfGrid = SDFGrid.createSVS()
sdfGrid.loadValuesFromFile(path="test_primitives.sdf")
sceneBuilder.addSDFGridInstance(
    sceneBuilder.addNode('SDFGrid', Transform(translation=float3(0, 0.6, 0))),
    sceneBuilder.addSDFGrid(sdfGrid, sdfGridMaterial)
)

The supported grid data structures are:

  • Normalized dense grid (NDG, which is a full mip hierarchy): sdfGrid = SDFGrid.createNDGrid(narrowBandThickness=2.5) (traversal is done on the SM)
  • Sparse voxel octree (SVO): sdfGrid = SDFGrid.createSVO() (traversal is done on the SM)
  • Sparse voxel set (SVS): sdfGrid = SDFGrid.createSVS() (uses the TTU, often the fastest method, but uses a lot of memory)
  • Sparse brick set (SBS): sdfGrid = SDFGrid.createSBS() (This is the best tradeoff in terms of performance and memory usage. Combines the TTU with traversal on the SM.)