Purr

It's a rust implementation of fogleman's primitive.

Created at Rusty Days Hackathon, 2020.

For more details about the algorithm, check this out: Hill Climbing

Feature

Simple rule, powerful result

Usage

Use this as lib

purrmitive = "*"

or as cli

cargo run --release --features=cli --bin=purr -- -i ./assets/input.png -o output.gif -n 100

most primitive flags are supported.

| Flag | Default | Description | | --- | --- | --- | | i | n/a | input file | | o | n/a | output file | | n | n/a | number of shapes | | m | 1 | mode: 0=combo 1=triangle 2=rect 3=ellipse 4=circle 5=rotatedrect 6=beziers 7=rotatedellipse 8=polygon(default 1) | | j | 0 | number of parallel workers (default uses all cores) | | r | 256 | resize large input images to this size before processing | | s | 1024 | output image size |

Example

using

-i ./assets/input.png -o output.gif -n 150

| primitive | input image | output image | process | | --- | --- | --- | --- | | triangle | | | | | ellipse | | | |

try it yourself for more.

Differences

Output Difference

All the graphs supported by primitive are implemented, and the output of all graphs (except the bezier curves) is basically the same.

The original primitive version of bezier curves supports line thickness adjustment, which is not possible in purr. But the result is still not bad.

Usage Difference

The key options and parameters of purr are the same as primitive, and in most cases you can just replace primitive with purr.

The rest of the parameters not listed above have not been implemented yet(these parameters will be supported in the future)

Performance Difference

All types of graphical fits are faster than primitive on my computer. See the next section for more specific data.

One possible factor is that in my implementation, the "next-step" buffer is skiped, by modified the diff function. It costs less comapring to copy the current buffer, update it then read it.

About Performance

This program is CPU intensive, it does all rendering in memory.

Purr is faster than the original implementation.

Benchmark using -n 100 -m {0-8}, see the results below.

hyperfine --parameter-scan m 0 8 -D 1 './target/release/purr -i ./assets/input.png -o assets/purr.{m}.png -n 100 -m {m}' '~/go/bin/primitive -i ./assets/input.png -o assets/primitive.{m}.png -n 100 -m {m}' --export-json benchmark.json

| Command | Mode | Mean [s] | Min [s] | Max [s] | result | |:---|---|---:|---:|---:|---:| |purr|combo| 10.256 ± 0.385|9.202|10.467| | |primitive|combo| 15.068 ± 0.213|14.785|15.405| | |purr|triangle| 6.927 ± 0.201|6.510|7.136| | |primitive|triangle| 10.577 ± 0.266|10.109|10.879| | |purr|rect| 6.224 ± 0.087|6.132|6.380| | |primitive|rect| 7.505 ± 0.105|7.388|7.765| | |purr|ellipse| 12.516 ± 0.098|12.354|12.674| | |primitive|ellipse| 14.643 ± 0.279|14.189|15.193| | |purr|circle| 15.054 ± 0.236|14.739|15.452| | |primitive|circle| 17.375 ± 0.187|17.156|17.760| | |purr|rotatedrect| 8.045 ± 0.081|7.938|8.220| | |primitive|rotatedrect| 9.181 ± 0.154|8.922|9.496| | |purr|beziers| 3.050 ± 0.127|2.815|3.223| | |primitive|beziers| 11.951 ± 0.173|11.563|12.114| | |purr|rotatedellipse| 12.282 ± 0.166|11.942|12.496| | |primitive|rotatedellipse| 27.795 ± 0.401|27.273|28.694| | |purr|polygon| 8.355 ± 0.219|7.988|8.748| | |primitive|polygon| 15.995 ± 0.275|15.426|16.431| |