Back to News
Advertisement
Advertisement

⚡ Community Insights

Discussion Sentiment

67% Positive

Analyzed from 375 words in the discussion.

Trending Topics

#here#best#compression#python#source#possible#streams#search#conversion#gifs

Discussion (12 Comments)Read Original on HackerNews

gsquaredxc•about 4 hours ago
> Here I should note that Python is not the best choice for CPU-bound software. I want to take the opportunity to learn Zig.

For optimizing CPU-bound operations in Python, there’s some low hanging fruit with numba. I would recommend this as a 5-minute solution to you limiting your algorithm because regular Python is too slow. I regularly tell people that if their Python program is slow enough to take several minutes, you could probably learn numba before it finishes.

sltkr•about 2 hours ago
PyPy is an option too.
davidkwast•about 3 hours ago
Nice advice
teddyh•8 minutes ago
For PNG, there’s ”pngcrush”: <https://pmt.sourceforge.io/pngcrush/>
BigTTYGothGF•about 1 hour ago
> We need to be careful here, because finding the actual best compression for an arbitrary format can be equivalent to solving the halting problem.

I don't see why that would be relevant here. The goal is to find the best compression of a given source among all possible DEFLATE streams, not to find the best compression of a given source among all possible ways to compress data. There are only so many DEFLATE streams shorter than the original source, you "simply" try them all, and then you've halted. That said:

> Now, Zopfli is the software that performs an exhaustive search across all possible syntactically valid streams

Maybe I don't understand LZ77 as well as I thought, but surely the search space is way too big to exhaust?

ghssds•18 minutes ago
> We need to be careful here, because finding the actual best compression for an arbitrary format can be equivalent to solving the halting problem.

That sentence was pasted unmodified from the LLM output.

dj_axl•30 minutes ago
Too big to exhaust... the concluding sentences of the article contain the details. "Hybrid search with pruning" is the current state of the codebase.
jjcm•about 2 hours ago
Neat. I tested it and here were my results:

ffmpeg conversion: 1,515 bytes, 0.05s

zgif conversion: 1,479 bytes, 90.1s

I used cursor to port it to rust as well, and it got the conversion time down to 20s. Still likely not worth it as even with the rust port it's a 400x increase in processing time (that scales exponentially) for a ~2% decrease in size.

Stitch4223•about 3 hours ago
Love the goal of supporting ancient browsers. What I’m missing in the article about images is images. For example the achieved results and some table of the amount of space saved using the compression methods described. Nonetheless interesting read.
richard_todd•about 2 hours ago
I agree, it would be way more interesting to see a table of original GIF sizes versus the improved size. I'm left wondering how bad the plain greedy version is on a typical image.
kernelbugs•about 4 hours ago
Cool effort! I appreciated the background into the why, as initially I thought this would have to do with optimizing animated GIFs and didn't realize there was a usecase for single frame GIFs anymore.
chrismorgan•about 4 hours ago
> didn't realize there was a usecase for single frame GIFs anymore

If the purpose is supporting NCSA Mosaic… I’m content to say that there isn’t. Definitely not “anymore”.