Pronay Peddiraju
Software Development
«
Procedural Image Generation using Markov-Chain Wave
Function Collapse
Procedural Content Generation (PCG) To improve on the existing models, my
technique used in games to ensure chain WFC (MkWFC), used Markov-chains
has always interested me as a design
randomness and replay-ability by
generating different maps or narratives in
game. Although most PCG algorithms rely
on noise-based techniques, I wanted to
pursue Wave Function Collapse (WFC),
which was an iterative constraint-based
content agnostic algorithm.
When performing my research on WFC,
to identify tile adjacency constraints from
input samples rather than using extensive
metadata files for the same. This allowed
me to reduce the design time from hours
to milliseconds by performing some
image processing steps. In doing so, my
implementation scaled with increasing
image sizes and constraints.
I came across two existing models of With the guidance of professors at SMU
Overlapping WFC model (OWFC) and the to the IEEE conference on games, and
WFC used to generate content — the
Tiling WFC model (TWFC). The OWFC
model generated accurate results, but
it had poor run-time performance and
lacked the ability to handle details. On the
other hand, the TWFC model provided
real-time performance and accounted
for detailed images but required hand-
curated metadata as input, which needed
substantial design time.
78
proposed implementation, called Markov-
SOFTWARE DEVELOPMENT
Guildhall, I was able to submit my research
I hope to publish my work. My intent is
to aid other like-minded developers to
incorporate WFC in their games with ease
and benefit from my algorithm.