[FFmpeg-devel] [RFC] Google Summer of Code 2013

[Maxim Polijakowski]

Am 27.02.2013 20:15, schrieb Michael Niedermayer:
> On Wed, Feb 27, 2013 at 07:29:38PM +0100, Maxim Polijakowski wrote:
>> Am 27.02.2013 16:55, schrieb Roger Pack:
>>> On 2/21/13, Lou Logan <lou at lrcd.com> wrote:
>>>> Hello,
>>>>
>>>> It's GSoC time again, so I'll get the ball rolling:
>>>> http://www.google-melange.com/gsoc/homepage/google/gsoc2013
>>>>
>>>> I started a new ideas list:
>>>> http://wiki.multimedia.cx/index.php?title=FFmpeg_Summer_of_Code_2013
>>> A few other wish list ideas, if anybody wants to run with them:
>> I suggest to remove the GoToMeeting entry from this list. It sounds
>> to me like another project that will never be completed in the GSoC
>> time frame (3 month). Let me explain why:
>>
>> - due to the fact the "decoder specification" has been created in
>> means of RE, it lacks the desired completeness. One need to dig into
>> the binary code in order to extract missing parts
>>
>> - there is two issues with the ELS coder - the heart of the coding
>> methods 2 and 3. Go2Meeting utilizes the augmented ELS that combines
>> the vanilla ELS with a some sort of Q-coder in order to increase
>> compression. This augmented version has been incompletely documented
>> by a software patent. No one in the world would be ever able to
>> reimplement this (weird) entropy coder using this patent alone.
>>
>> - fortunately, there is an open-source implementation of this coder
>> written by the author of this compression method, Mr. Douglas
>> Whiters. Though this code has been released under the GPL and a
>> relicensing is required in order to use it within FFmpeg. One need
>> to contact the author and ask for the appropriate permission. Should
>> he disagree, we'll end up having another non-free piece of code.
>>
>> - FFmpeg's implementation of the JPEG need to be extended to include
>> tables for Go2Meeting. The attached spec doesn't describe these
>> tables.
> quantization and or huffman tables ?
> are they stored in the files or constant ?
>
> if constant and a encoder is available, you can encode the output of
> -f lavfi mptestsrc
> the values for the tables can then be inferred from the generated
> bitstream.
> The alternative is to find the tables in the binary which may or may
> not be easier

Thank you for the suggestions. The tables are constant and I've already 
extracted those from the binary because it's just plain libjpeg6b...

I just wanted to say that the specification made by Kostya (great job 
anyway!) is not sufficient enough to write the decoder from scratch. 
Such "extra" knowledges are essential in order to be able to finish the 
GSoC task...

Best regards
Maxim