#mcmc #image-processing #computer-vision #graphics #generative-imaging

convchain

Bitmap generation from a single example with convolutions and MCMC

3 unstable releases

0.2.1 Aug 11, 2021
0.2.0 Aug 11, 2021
0.1.0 Aug 10, 2021

#1419 in Game dev

MIT license

21KB
458 lines

ConvChain (Rust Port)

A port of Maxim Gumin's ConvChain to Rust:

ConvChain is a Markov chain of images that converges to input-like images. That is, the distribution of NxN patterns in the outputs converges to the distribution of NxN patterns in the input as the process goes on.

Please have a look on the original repo for a more thorough description of the application of the Metropolis algorithm to the problem.

fn main() {
    // Generate a sample (this is the SimpleMaze example):
    let pattern = [
        true,  true,  true,  true,
        true, false, false, false,
        true, false,  true, false,
        true, false, false, false,
    ];
    let sample = ConvChainSample::new(&pattern, 4, 4);

    // Initialize the chain using given sample and a
    // - 32x32 output size
    // - receptor size of 2
    // - temperature of 1.0
    let mut chain = ConvChain::new(&sample, 32, 2, 1.0);
    
    // Generate the 32x32 field using 10 iterations.
    let field: &[bool] = chain.process(10);
}

Example run

Here are a couple of outputs generated with the code, using different values for the receptor size and number of iterations:

Input r=3, it=10 r=5, it=10 r=5, it=20 r=5, it=100

This repository provides ports of both original "slow" and "fast" implementations. To run the example, execute the following from the repository root:

$ cargo run --release --example fast

This will process the jobs defined in resources/samples.xml and produce output images in the current directory.

To run the benchmarks, execute

$ cargo bench

Dependencies

~310KB