[FFmpeg-devel] Issues with ljpeg

[Mathieu Malaterre]

On Fri, Aug 15, 2008 at 11:40 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Fri, Aug 15, 2008 at 07:31:13PM +0200, Mathieu Malaterre wrote:
>> A couple of comments on ljpeg encoder:
>>
>> 1. (minor) There is no such thing as Ss=8 in JPEG lossless.
>>
>>  $ ffmpeg -f image2 -i beach.jpg -f image2 -pred 7 -vcodec ljpeg
>> /tmp/out0.ljpeg
>>
>> As far as I under the code when pred is > 7 then predictor=7 is used
>> (see PREDICT macro in mjpeg.h). But Ss=8 is stored in the jpeg header.
>>
>
>> 2. A more serious issue is for pred=4, pred=5 and pred=6. I made some
>> snapshots of what I can see using IJG decompressor (+lossless patch)
>> here:
>
> what is "IJG decompressor (+lossless patch)" ?

Independent JPEG Group (ijg.org)
lossless patch (Ken Murchison): ftp://ftp.oceana.com/pub/ljpeg/


>>
>> http://jpeg.sourceforge.net/ffmpeg/
>>
>>   Using the following command lines:
>>
>> ./ffmpeg -f image2 -i beach.jpg -f image2 -pred 6 -vcodec ljpeg  /tmp/out6.ljpeg
>> ./ffmpeg -f image2 -i beach.jpg -f image2 -pred 5 -vcodec ljpeg  /tmp/out5.ljpeg
>> ./ffmpeg -f image2 -i beach.jpg -f image2 -pred 4 -vcodec ljpeg  /tmp/out4.ljpeg
>>
>> Sorry I could not find the issue, but if anyone understand where the
>> issue could be coming from, I volunteer to prepare the patch.
>
> look at the int8_t casts and &0x1FF
> i think these are wrong, patch welcome and dont forget
> 1. the regression tests, they might need to be updated
> 2. check samples.mplayerhq.hu that the change does not break any ljpegs from
>   there

Thanks. Will do.


-- 
Mathieu