DM355 mpeg4enc artifacts

Robin Iddon

I'm getting encoding artifacts with the latest DM355 MPEG-4 encoder with the latest DVSDK that I didn't get with older codec versions with the older v2 DVSDK.

I've provided a detailed description of the problem with example MPEG-4 streams and stills of the artifacts on this web page.

Regards,

Chris Paulson-Ellis <chris@edesix.com>

over 13 years ago

0 Akshay Panday over 13 years ago

Expert 1640 points

Hi Chris / Robin

Can you try repeating this test with the quantization parameter (Qp) capped at 4?

We have observed a similar / related issue where low quantization value can cause an overflow in one of the internal buffers resulting in errors.

Alternately, you can upgrade to release 2.00.00.04 where this is taken care of internally.

Regards,

Akshay

0 Robin Iddon over 13 years ago in reply to Akshay Panday

Intellectual 380 points

Hi Akshay,

Thanks, I'll try that in a few days (I'm on Holiday until Tuesday). Where can I download the 2.00.00.04 release? I can only see 2.00.00.02 here:

http://software-dl.ti.com/dsps/dsps_public_sw/codecs/DM355/index_FDS.html

Chris.

0 Akshay Panday over 13 years ago in reply to Robin Iddon

Expert 1640 points

Chris,

I'm not sure why it's not yet available on the download server. Perhaps your local TI rep might help you in getting a copy.

Let me know if the workaround helps.

Regards,

Akshay

0 Robin Iddon over 13 years ago in reply to Akshay Panday

Intellectual 380 points

Akshay,

I tried the work-around, but I'm still getting artefacts. I wasn't using the codec-specific dynamic parameters before, so perhaps I did something wrong. I'm using these parameters now:

VIDENC1_Params:
    size                    sizeof(VIDENC1_Params)
    encodingPreset          XDM_HIGH_QUALITY
    rateControlPreset       IVIDEO_STORAGE
    maxHeight               height
    maxWidth                width
    maxFrameRate            25000
    maxBitRate              2000000
    dataEndianness          XDM_BYTE
    maxInterFrameInterval   1
    inputChromaFormat       XDM_YUV_422ILE
    inputContentType        IVIDEO_PROGRESSIVE
    reconChromaFormat       XDM_DEFAULT

VIDENC1_DynamicParams:
    size                    sizeof(IMP4VENC_DynamicParams)
    inputHeight             width
    inputWidth              height
    refFrameRate            25000
    targetFrameRate         25000
    targetBitRate           2000000
    intraFrameInterval      13
    generateHeader          XDM_ENCODE_AU
    captureWidth            0
    forceFrame              FORCE_FRAME_NONE
    interFrameInterval      0
    mbDataFlag              0

IMP4VENC_DynamicParams:
    intraAlgo               IMP4VENC_INTRA_INTER_DECISION_LQ_HP
    numMBRows               height / 16
    initQ                   0
    rcQMax                  4
    rcQMin                  1
    intraFrameQP            0
    interFrameQP            0
    rateFix                 0
    rateFixRange            0
    meAlgo                  IMP4VENC_ME_HQ_LP
    skipMBAlgo              IMP4VENC_SKIP_MB_LQ_HP
    unrestrictedMV          IMP4VENC_UMV_LQ_HP
    mvDataEnable            0

In parallel, I'll try and get the latest release too.

Chris.

0 Akshay Panday over 13 years ago in reply to Robin Iddon

Expert 1640 points

Chris,

Your rcQMin should be 4 and rcQMax can be anything upto 31.

Regards,

Akshay

0 Robin Iddon over 13 years ago in reply to Akshay Panday

Intellectual 380 points

Akshay,

Ah - I got it backwards!

This is definitely a big improvement. With rcQMin=4 & rcQMax=31, I've only encountered 1 artefact in about 30mins of footage reviewed so far.

Chris.

0 Robin Iddon over 13 years ago in reply to Robin Iddon

Intellectual 380 points

Akshay,

I managed to get hold of release 02_00_00_04 of mpeg4enc. I'm still getting artifacts frequently enough for it to be a problem. The release seems to behave in the same way as the previous version with rcQMin=4, but without needing to apply that work-around.

Are there any plans for further releases with more fixes? I could just about get away with shipping product with this release, but only if I can promise an update to any customers that complain.

Chris.

0 Akshay Panday over 13 years ago in reply to Robin Iddon

Expert 1640 points

Chris,

Well, essentially that is what the 2_00_00_04 release does internally. I'm afraid you might have to experiment with increasing the min Q further to see whether it passes your quality criteria.

According to our analysis, it appears that this issue crops up in areas of high entropy or very sharp edges (like overlay text) when some of the internal buffers overflow. Although we support these codecs for minor bug fixes and workarounds, this issue requires significant redesign of the codec internals and we currently do not have plans for that level of support.

Regards,

Akshay

0 Robin Iddon over 13 years ago in reply to Akshay Panday

Intellectual 380 points

Hi,

With the rcQMin=8 and rcQMin=31 we still see these artifacts from time to time and not in any way that is predictable.

Are there other parameters that should be adjusted also?

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Hope you have tried with rcQMax = 31 and rcQMin = 8 , not rcQMin=31

Regards, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

In all of our tests rcQMax=31.

We tested with a variety of rcQMin values including 0,3,8 and 31 and in all cases the artefacts appear.

Regards,

Robin

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Is it possible to share the part of the stream with rcQMin = 8 and rcQMax = 31 where the artefacts are still seen?

Regards, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

I take it you want the MPEG4 stream rather than the individual frames decoded?

We can extract the MPEG4 stream that corresponds to the frame sequence linked to above and upload it to the same place.

Regards,

Robin

P.S. it turns out that we don't have that data for the original images so we'll capture another example and post that up - hopefully in the next day or two.

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Yes, we want to have a look at MPEG-4 stream not the decoded images.

Regards, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

This MPEG-4 stream was recorded with rcQMin=8, rcQMax=31. It is 15 mins long (25fps) and has 4 artefacts at the following frames (as burned in to the frames - first frame is 0):

5058 - 5069
7307 - 7318
12572 - 12583
15406 - 15417

Regards,
Chris.

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

We are analyzing stream to understand possible causes for this behavior. Will get back to you if we have any suggestions.

Regards, RSV

0 Robin Iddon over 13 years ago in reply to Robin Iddon

Intellectual 380 points

Hi,

Did you manage to download the file and analyze it?

http://www.edesix.com/downloads/mpeg4enc-err/rcqmin-8-rcqmax-31.mpeg4

Best wishes,

Robin

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Hi,

Sorry for the delay in response. We have analyzed the stream but found any possible issues.

I am not sure any further steps here..

Regards, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

I am not sure I understand what you mean.

Are you saying you've analyzed it and found no issues? Clearly the pink blobs weren't really there so they must have been an issue in the CODEC.

The old CODEC did not behave this way, so one step would be to port the old CODEC into the new framework ...

Best wishes,

Robin

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Hi,

The pink blocks are there, what I meant is we are not able to identify possible cause for the same from steam properties/attributes.

What is the version of old codec which doesn't have this issue?

Regards, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

Hi,

As I said in the web page linked from my original post: The older codec from DM355 codec package 1_14_002 used with DVSDK 2_00_00_22 does not show the problem.

Regards, Chris.

0 RSV Prasad over 13 years ago in reply to Robin Iddon

Intellectual 385 points

Hi,

DM355 codec package 1_14_002 -- doesn't really refer to particular DM355 MPEG-4 encoder version. Can you please check the release notes of MPEG-4 encoder of that package and let me know Build ID mentioned there?

Regars, RSV

0 Robin Iddon over 13 years ago in reply to RSV Prasad

Intellectual 380 points

Hi,

The release notes say Version 1.14, March 2010, Build ID: 1.14.002

Chris.

0 Robin Iddon over 13 years ago in reply to Robin Iddon

Intellectual 380 points

Is this issue going to be addressed further?

Would $$$ help - I am sure something can be arranged?

We need this fixed.

I cannot believe that TI doesn't need this fixed.

Best wishes,

Robin Iddon

EDESIX Limited

robin@edesix.com

Processors

Processors forum

DM355 mpeg4enc artifacts