Genetic Image Evolution in Rust

A multithreaded genetic algorithm written in Rust that reconstructs target images using overlapping semi-transparent polygons and ellipses. Over successive generations, individuals mutate and cross over to approximate the target image pixel-by-pixel.

---

Examples

Target (images/)	Reconstructed Output (Root)
Original: Charles Darwin	Evolved (~96.7% fitness)
Original: 9c	Evolved (~95.0% fitness)

---

Features

• Parallel Evaluation: Scales across CPU threads with rayon for concurrent rasterization and fitness scoring.
• Vector Rasterization: Uses tiny-skia for rendering semi-transparent paths and shapes directly into memory buffers.
• Dynamic Hyperparameters: Automatically scales mutation steps, shape bounds, and survival thresholds as overall population fitness increases.
• Shape Variety: Combines variable-vertex polygons with bounding ellipses to capture both broad gradients and sharp edges.

---

Getting Started

Prerequisites

Ensure you have a recent Rust toolchain installed:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

Running

Execute with native CPU vectorization enabled in release mode:

RUSTFLAGS="-C target-cpu=native" cargo run --release -- <image_file> <generations> <population_size>

Examples:

# Evolve darwin.png with 9000 generations and 300 individuals
RUSTFLAGS="-C target-cpu=native" cargo run --release -- darwin.png 9000 300

# Evolve 9c.png with custom generation limits
RUSTFLAGS="-C target-cpu=native" cargo run --release -- 9c.png 10000 500

Arguments

Parameter	Default	Description
image_file	9c.png	Name of the file inside images/
generations	1000	Total generations to run
population_size	350	Number of candidate individuals per generation

---

Outputs

Final renders are written to the out/ directory with run parameters and the final fitness score in the filename:

out/{file_name}_{generations}_{pop_size}_{chromosome_count}_{fitness}.png