Hacker News Reader

Comment by cornstalks | Hacker News Reader

Fabrice Bellard (the wizard behind FFmpeg, QEMU, etc.) has his own entry, which I find amazing: a ~4 kB program that outputs a ~50 kB image (of Lena), in addition to being able to decompress other images. His entry's hint page: http://www.ioccc.org/2018/bellard/hint.html

View on HN

Here's the default output (embedded 1220 byte 128x128 Lena picture) of Fabrice's IOCCC entry converted into PNG:

https://pasteboard.co/Hk60KEv.png

Of course there are plenty of compression artifacts, but very impressive for just 1220 bytes of image data embedded in decompressor written in C totaling only 3984 bytes of source code!

No idea about what kind of service pasteboard.co is, hopefully the image will stay up for a while...

Edit: For comparison, here's a roughly corresponding original Lena image converted into 1269 byte 128x128 JPEG.

  convert -resize 128x128 -strip -quality 12% lena.png lena.jpg

https://pasteboard.co/Hk67yXC.png

> No specific obfuscation was needed as the algorithms already have a significant complexity.

Hehe...

He wasn't joking: http://www.ioccc.org/2018/bellard/prog.c

Funny, the compiler seems to have a hard time with it. Running it OTOH is instant.

It's a 3KB decoder and 1KB of image data, from the hint: "The actual image data is 1220 byte long, which gives a compression ratio of 40"

Can you explain a bit more how this works? I am not familiar at all with these challenges

50kB to 4kB is a 10x compression, it doesn't seem that huge, what's the achievement here?

Achieving 10x compression in a 4kB program+data file hybrid is pretty hard to do.

From the author's comments:

"The uncompressed image is 12 times larger than the source code of the program which includes the image data and the complete decoder. The actual image data is 1220 byte long, which gives a compression ratio of 40. Using a JPEG-like algorithm would not be enough to reach this level of compression (the Lena image would be barely recognizable). So the algorithm is based on the latest advances in image compression."

http://www.ioccc.org/2018/bellard/hint.html

From the author’s notes:

> The actual image data is 1220 byte long, which gives a compression ratio of 40.

Those 4k include the decoder source code and the C string representation of the image data

For those using iTerm, a nice way to try this out is running it through imgcat:

  make && ./prog < lena512.bin | imgcat

If you have ImageMagick installed the `display` utility can also show PPM images from stdin:

    ./prog < lena512.bin | display

---

btw, the `animate` program can also be used to display the animated GIFs from endoh1.

Is there a program available somewhere to generate our own compressed image (.bin) files?

GIMP?

Didn't he come up with a custom image compression scheme? Or am I understanding this wrong?

It might be custom. I haven't checked or compared. There are options to save images in other formats with GIMP.

You understood correctly. This is a custom format that has some ideas from h.264 and who knows what else.

Almost certainly not yet. The compression method seems to be pretty advanced, borrowing ideas from h.264 etc. and is almost certainly incompatible with any currently existing format.

If we're lucky, Bellard might eventually post an image compressor program on his website at https://bellard.org/.

Reply on news.ycombinator.com