Development Log - VR Archaeology

cjwidd · Jun 2018

DISCLAIMER

This is a development blog to document and record progress as a part for an ongoing research program. However, please feel free to provide comments, criticisms, tips / tricks, and other relevant insights - I appreciate any advice you can offer. Thanks!
_______________________________________________________________________________________________________________________

Objective: create game ready VR environment to support a variety of archaeological tasks
Deadline: September, 2018

Image: https://us.v-cdn.net/5021068/uploads/editor/vh/kt4g03dtjkbc.png

Experimenting with material creation in Substance Designer
Material inspired by Lower Antelope Canyon and an eroded desert cliff material by Daniel Thiger

Image: https://us.v-cdn.net/5021068/uploads/editor/up/8d2ilobrqccy.png

Quick Zbrush sculpt of primary play space
Added dirt / piles
Test proportions

Image: https://us.v-cdn.net/5021068/uploads/editor/j5/sftwgu40flsa.png

Retopologize high poly mesh to low poly variant

Image: https://us.v-cdn.net/5021068/uploads/editor/y0/ffo533xtgs0n.png

Bake mesh maps in Substance Painter
Check for artifacts; looks OK

Image: https://us.v-cdn.net/5021068/uploads/editor/yn/ug9fpph9gt0x.png

Image: https://us.v-cdn.net/5021068/uploads/editor/5g/r4mlnnd1hmtv.png

Material renders for candidate surfaces; displacement maps seem fine; good color

cjwidd · Jun 2018

Objective: Begin work in UE4

[video] Testing uniform displacement
[video] T esting displacement with spherical mask
[video] T esting displacement with spherical mask and texture set
[video] A dded AO slider
added static switch parameter to toggle grayscale or channel packed texture (see below)

Image: https://us.v-cdn.net/5021068/uploads/editor/ha/weiua2tdoqgo.png

*experiments include displacement blend materials by Lincoln Hughes

Notes:
I heard tessellation was an issue in UE4, but coming face to face with it was another thing. The implementation clearly works, and I'm sure there are settings that could be managed to produce a usable end result, but I'm questioning whether it is worth it.

Alternatives might include static tessellation instead of dynamic (camera based). In doing so, a spherical mask could be defined based on a static coordinate, and then just tessellate the ground plane that way. It seems the engine is happier when the tessellation doesn't have to update constantly (of course), however, I have not seen any realtime projects for game-ready environments that are using it successfully - that's not to say they don't exist, I just haven't seen them personally.

Cons:

Texture popping, camera stutter
Seams on modular pieces
Extra shader complexity to manage subdivision levels

Pros:

Project probably would not benefit from a modular kit
Managing geometry inside the Editor
Managing texturing inside the Editor with World Space UVs

Question

To tessellate, or not?

cjwidd · Jun 2018

Objective: average displacement to +/- range, rather than + only (*tip from Jan Kaluza)

Image: https://us.v-cdn.net/5021068/uploads/editor/3p/s6eo4o8629fy.png

Image: https://us.v-cdn.net/5021068/uploads/editor/s2/wmutpghh2d7g.png

Image: https://us.v-cdn.net/5021068/uploads/editor/9n/qsnyw893q6lb.png

Notes:

Using a .jpg map in the height channel produced obvious displacement artifacts and created an offset into the -z axis, creating a gap between the user's feet and the ground plane. Using an .exr version of the same map produced no artifacts, and no gap either. But why?
I assume it is related to differences in bit depth or dynamic range of the images.

cjwidd · Jun 2018

Objective: vertex paint on tessellated static mesh

Testing vertex paint utilities in Editor on subdivided base mesh (~8k verts). On a highly subdivided mesh, there are sufficient vertices to support detail vertex painting in the editor.
However, when testing on a low poly model that was tessellated in the shader, the vertex painting utilities fail. Apparently, vertex painting only respects base mesh vertices, but not tessellated vertices. Presumably this has to do with dynamic tessellation options which constantly update the subdivision levels based on camera position, etc.

Image: https://us.v-cdn.net/5021068/uploads/editor/ef/q7ffyx0i8et3.png

Applying same material to scaled landscape actor.
After clamping LOD system, tessellation occurs in predicted location, but efficiency of subdivision fails (see above)

*Thanks to @radiancef0rge for input

cjwidd · Jun 2018

Image: https://us.v-cdn.net/5021068/uploads/editor/jk/3elr948bjojp.png

Image: https://us.v-cdn.net/5021068/uploads/editor/dr/7lt2nadwt5py.png

Image: https://us.v-cdn.net/5021068/uploads/editor/mf/ok2hdc64cemg.png

This is a terrain shader (see first image) with a variety of features
In the second image, (on the right) you can see several global features and the contents of one layer - there are two more layers.
Each layer includes a texture sampler for albedo (diffuse), normal, roughness, height (displacement), and roughness
There are parameters to control roughness contrast, and a multiplier to adjust roughness based on that contrast
There are parameters to control the intensity of displacement for each height channel; the shader will accept either greyscale displacement maps, or channel packed (red) displacement maps - Megascans exports some displacement in the red channel
The entire shader makes use of world space UVs and has parameters to control tiling and UV rotation per layer (*thanks @lincolnhughes). The shader will support vertical or horizontal tiling. The benefit of world space UVs is that regardless of how the terrain object is moved around in world space, the textures will always be projected to a unique position in world space, e.g. easier to modify terrain without having to worry about UVs.
There is a unique tessellation modifier appended at the end to control a point of origin for tessellation and a mask. After doing some experiments, from a view height of about 1.8m, it's difficult to ascertain any benefits of added tessellation for objects after about 18m, and that is with a very critical eye - in VR it will probably be even shorter.
The radius / hardness of the tessellation mask can be modified, and the tessellation intensity as well. Currently the tessellation mask is defined by a static point in space, but that point can be modified freely; the reason for this is because I was experiencing hiccups on my machine (single frame drops) using dynamic tessellation.
The shader supports 3-layer blends (weight-blends) based on vertex color. In this way, we can make use of UE4's terrain painting tools to texture the landscape without ever jumping into a 3rd party texturing software; provided there are sufficient vertices to paint on.
You can see an example of this weight-blending occurring on the terrain (see third image). Even with very dissimilar texture (tile vs. sand / rock), a smooth blend can be achieved.