add qoi-rs benchmark

[Joshix-1]

Made my one library because I'm not in the numpy ecosystem and wanted an alternative using more normal python datatypes. Performance seems to be similar.

Test image	Method	Format	Input (kb)	Encode (ms)	Encode (kb)	Decode (ms)	SSIM
all black ('best' case)	PIL	jpg @ 80	6075.0	2.26	32.5	2.52	1.00
all black ('best' case)	PIL	png	6075.0	16.44	6.0	10.51	1.00
all black ('best' case)	opencv	jpg @ 80	6075.0	2.10	32.5	1.97	1.00
all black ('best' case)	opencv	png	6075.0	8.82	7.7	12.65	1.00
all black ('best' case)	qoi	qoi	6075.0	2.08	32.7	1.23	1.00
all black ('best' case)	qoi-lossy-0.50x0.50	qoi	6075.0	0.92	8.2	0.92	1.00
all black ('best' case)	qoi_rs	qoi	6075.0	1.91	32.7	0.88	1.00
koi photo	PIL	jpg @ 80	6075.0	3.47	274.6	5.34	0.94
koi photo	PIL	png	6075.0	638.59	2821.5	35.79	1.00
koi photo	opencv	jpg @ 80	6075.0	2.99	275.3	3.62	0.94
koi photo	opencv	png	6075.0	44.85	3121.5	21.84	1.00
koi photo	qoi	qoi	6075.0	13.93	3489.0	9.40	1.00
koi photo	qoi-lossy-0.50x0.50	qoi	6075.0	3.85	980.1	2.86	0.95
koi photo	qoi_rs	qoi	6075.0	10.78	3489.0	7.34	1.00
random noise (worst case)	PIL	jpg @ 80	6075.0	6.46	1351.6	11.66	0.55
random noise (worst case)	PIL	png	6075.0	160.09	6084.5	21.63	1.00
random noise (worst case)	opencv	jpg @ 80	6075.0	6.13	1351.7	9.75	0.55
random noise (worst case)	opencv	png	6075.0	29.66	6086.9	7.45	1.00
random noise (worst case)	qoi	qoi	6075.0	8.46	8096.0	4.73	1.00
random noise (worst case)	qoi-lossy-0.50x0.50	qoi	6075.0	2.54	2019.1	2.23	0.24
random noise (worst case)	qoi_rs	qoi	6075.0	7.94	8096.0	3.36	1.00

[kodonnell]

Cool! What OS are you running on? Yours was slightly faster on those tests, whereas when I run it (WSL2 Windows 11 i7-12700H, conda environment) it's slightly slower as below. (EDIT: sorry, only skimmed table. Looks similar, with yours generally having a slight edge)

Test image	Method	Format	Input (kb)	Encode (ms)	Encode (kb)	Decode (ms)	SSIM
all black ('best' case)	PIL	jpg @ 80	6075.0	3.49	32.5	4.08	1.00
all black ('best' case)	PIL	png	6075.0	18.20	6.0	5.20	1.00
all black ('best' case)	opencv	jpg @ 80	6075.0	2.49	32.5	2.62	1.00
all black ('best' case)	opencv	png	6075.0	9.75	7.7	8.23	1.00
all black ('best' case)	qoi	qoi	6075.0	3.00	32.7	1.94	1.00
all black ('best' case)	qoi-lossy-0.50x0.50	qoi	6075.0	1.23	8.2	1.25	1.00
all black ('best' case)	qoi_rs	qoi	6075.0	3.14	32.7	0.97	1.00
koi photo	PIL	jpg @ 80	6075.0	5.92	274.6	8.41	0.94
koi photo	PIL	png	6075.0	700.26	2821.5	53.39	1.00
koi photo	opencv	jpg @ 80	6075.0	4.43	275.3	6.80	0.94
koi photo	opencv	png	6075.0	57.41	3121.5	33.62	1.00
koi photo	qoi	qoi	6075.0	17.96	3489.0	12.55	1.00
koi photo	qoi-lossy-0.50x0.50	qoi	6075.0	5.55	980.1	3.46	0.95
koi photo	qoi_rs	qoi	6075.0	15.91	3489.0	10.48	1.00
random noise (worst case)	PIL	jpg @ 80	6075.0	9.17	1351.5	15.91	0.55
random noise (worst case)	PIL	png	6075.0	148.40	6084.5	38.38	1.00
random noise (worst case)	opencv	jpg @ 80	6075.0	9.51	1351.5	13.91	0.55
random noise (worst case)	opencv	png	6075.0	34.48	6086.9	9.71	1.00
random noise (worst case)	qoi	qoi	6075.0	12.73	8096.0	5.40	1.00
random noise (worst case)	qoi-lossy-0.50x0.50	qoi	6075.0	3.50	2019.1	2.46	0.24
random noise (worst case)	qoi_rs	qoi	6075.0	12.82	8096.0	4.57	1.00

Re merging, I was going to but had a look at https://github.com/phoboslab/qoi and there are 3 other python libs (not including yours), so if I was going to include yours I should probably include theirs. Thoughts?

[Joshix-1]

I'm running on native Arch Linux.

Aren't the other libraries written in python? I doubt they have similar performance. I don't really care if you merge this, or add the others. I just was curious and added my library to your benchmark, and thought I could share it.