Fevh264 - simple baseline h.264 encoder

[imcold]

Fevh264 is a very simple baseline h.264 encoder. I'm releasing it in hope that somebody finds it useful if he decides to learn about h.264 encoding process and as a base for experiments; it's not a serious competition to "real" h.264 encoders. It's written in Pascal (Freepascal) with some hand-written assembly and shouldn't be hard to understand.

It uses the command line interface (see fevh264_cli -h for switches). Input is restricted to raw YUV 4:2:0 (width and height requied), yuv4mpeg or Avisynth scripts; output is raw .264 (AnnexB) file.

Update:
-added simple ratecontrol for 2 pass avg. bitrate encoding
-added multiref support
-GUI frontend with visualization support, on-the-fly QP change support. Screenshot (windows, linux):
http://i266.photobucket.com/albums/ii260/imcoldz/prog/983027c1.png http://i266.photobucket.com/albums/ii260/imcoldz/prog/th_236842f5.png (http://s266.photobucket.com/albums/ii260/imcoldz/prog/236842f5.png)

Limitations:
-won't work on cpu without sse2 support, also the asm is 32bit only
-can't be build as a library currently
-all files set to 25 fps

Missing useful baseline stuff, to be added later:
-inter mb partitions
-deblock

Sample (parkjoy, using older version of encoder): http://www.mediafire.com/?cyk53kqx88tcc10

Source, binaries: https://sourceforge.net/projects/fevh264/

Compilation:
Win32 and Linux 32bit are supported. You'll need Yasm and Freepascal 2.x + Lazarus IDE.
1. go to frontend subdirectory, run build_asm.bat/.sh to compile asm files
2. open fevh264_cli.lpi / fevh264_gui.lpi through Lazarus IDE and build

If you find any bugs (crashes, unplayable output files), let me know.

[Terranigma]

Not bad. Not bad at all. I just wish it could handle non mod-16 resolutions.

[lnatan25]

Haha, Pascal. But it's commendable, good stuff!

[imcold]

Support for non-mod16 resolution added.
Haha, Pascal.
Yes, still alive and well, though fallen out of favor.

[TheImperial2004]

Great stuff indeed . ;)

Why don't you write this encoder for GPU (e.g. CUDA) from the very begining ? It will be great and easier than adding it later when you have a big source tree :)

All the best ...

[imcold]

1. there won't be a big source tree, that's one of the goals - any gpgpu efforts should be aimed on x264
2. from what I've seen so far, usefulness of gpgpu in video encoding is very limited
3. I don't have gpgpu capable graphic card (5+ years old)

[TheImperial2004]

1. there won't be a big source tree, that's one of the goals - any gpgpu efforts should be aimed on x264
2. from what I've seen so far, usefulness of gpgpu in video encoding is very limited
3. I don't have gpgpu capable graphic card (5+ years old)

Neverthless , I'll keep my eye on this project . I might even develop a simple GUI for it if I had enough time , And I'll consult you on the matter ;)
Good luck :)

[imcold]

I had some time recently so I made some improvements to the encoder and made a new encoding GUI, so you can observe the encoder in action: it displays macroblock types, motion vectors and some statistics about currently encoded frame and it allows you to change QP on the fly. Some settings are non-functional/not available at the moment, though: 2pass, chroma qp changing. Have fun.

[benwaggoner]

Yes, still alive and well, though fallen out of favor.
Pascal was the last commonly-used programming language I felt adequately good at :). Got a 5 on my AP CS test... in Pascal... in 1988...

Did you pick Pascal because you liked it, or for another reason?

Also, can you give some background on the origin and goals of this project?

[imcold]

I generally prefer using (Free)pascal instead of C/C++. Origin - I got interested in data compression, especially video compr. way back and started making progs using different compression algos. Things evolved over the years and voila. Goal - learn about h.264 standard specifics and how encoders work in general, and I guess learning by implementing it is a very good way to do this.