This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

OpenCV on TMS320DM6446

Other Parts Discussed in Thread: TMS320DM6446, DM3730, OMAP-L138, OMAP3530

Hello,

I need to port OpenCV 2.1 tracking application from Windows XP PC on DSP TMS320DM6446. How can I  start? Where we can get build environment fro OpenCV DSP version?

How can I convert floating point stuff to fixed point?

I would greatly appreciate your help in this issue. Many thanks in advance.

Best Regards,

  Vladimir

  • Vladimir,

    We have an experimental base port of OpenCV 1.x  and OpenCV 2.3 on the C674X floating point DSP. We also have a version of OpenCV 1.x for C64x+ DSP but we wil have to check if they have all the APIs that may be required for your tracking Application are part of that version. The version we have is not hand optimized for the fixed point core but a version that is cross compiled with optimized Compiler setttings for the DSP core and some changes to bypass OS specific stuff and to resolve memory alignment issues.

    For you to optimize the floating point code for fixed point processor, you can use Q point math in your code using IQMAth library or optimize floating point math operations by leverage TI`s C64x FastRTS library as a starting point.

    Regards,

    Rahul

     

  • Hello,Rahul,

    We have an experimental base port of OpenCV 1.x  and OpenCV 2.3 on the C674X floating point DSP. 

    We also have a version of OpenCV 1.x for C64x+ DSP

    Could you tell me where the ported OpenCV is?

    Thanks.

  • Hello Rahul,

     Many thanks for your answer. How can I get OpenCV 1.x for C64x+ DSP ? As you said,  I need to check APIs in order to see that we can use it.

    I'll look IQMath and fastRTS, can you post a link here?

    Best Regards,

      Vladimir

  • Vladimir,

    We cannot distribute the library directly on the forums, so could you send us an email on the developer mailing list speciifed on the wiki so that I can setup a FTP to share the C64x+version of OpenCV 1.x.

    http://processors.wiki.ti.com/index.php/Software_libraries#Developer_Mailing_List

    You can find the IQMATH and fastRTS for C64x from the following links:

    IQMATH: http://www.ti.com/tool/sprc542

    C64x fastRTS: http://www.ti.com/tool/sprc122 

     

    Regards,

    Rahul

  • Rahul Prabhu said:

    Vladimir,

    We cannot distribute the library directly on the forums, so could you send us an email on the developer mailing list speciifed on the wiki so that I can setup a FTP to share the C64x+version of OpenCV 1.x.

    http://processors.wiki.ti.com/index.php/Software_libraries#Developer_Mailing_List

    You can find the IQMATH and fastRTS for C64x from the following links:

    IQMATH: http://www.ti.com/tool/sprc542

    C64x fastRTS: http://www.ti.com/tool/sprc122 

     

    Regards,

    Rahul

    Hello Rahul,

    Can I get a copy?

    Thank.

  • Rahul Prabhu said:

    Vladimir,

    We have an experimental base port of OpenCV 1.x  and OpenCV 2.3 on the C674X floating point DSP. We also have a version of OpenCV 1.x for C64x+ DSP but we wil have to check if they have all the APIs that may be required for your tracking Application are part of that version. The version we have is not hand optimized for the fixed point core but a version that is cross compiled with optimized Compiler setttings for the DSP core and some changes to bypass OS specific stuff and to resolve memory alignment issues.

    For you to optimize the floating point code for fixed point processor, you can use Q point math in your code using IQMAth library or optimize floating point math operations by leverage TI`s C64x FastRTS library as a starting point.

    Regards,

    Rahul

     

    Hi Rahul,

    1) Will there be a version of OpenCV 2.3 for C64X+ DSP ?

    2) When I use IQMath or FastRTS to optimise floating point OpenCV code for fixed point C64X+, does it mean that

        I have to use IQMath or FastRTS functions to convert every floating point computation in OpenCV code ?

        This can be error-prone and tedious ?

    Thanks,

    Mr Tan

  • Hi Rahul,

    I sent the email for this address. Could you sned me the FTP link?

    Many thanks in advance,

    Best Regards,

    Vladimir

  • VLadimir,

    I have sent you the ftp site link and the instructions to access the source. To rebuild these libraries please use the latest C6000 compiler version for optimal performance.

    Regards,

    Rahul

  • Rahul,

     You probably sent it to my work address, could you also send it to shcherb@gmail.com.

    Sorry for inconvinience,

     Best Regards,

     Vladimir

  • Rahul Prabhu said:

    Vladimir,

    We have an experimental base port of OpenCV 1.x  and OpenCV 2.3 on the C674X floating point DSP. We also have a version of OpenCV 1.x for C64x+ DSP but we wil have to check if they have all the APIs that may be required for your tracking Application are part of that version. The version we have is not hand optimized for the fixed point core but a version that is cross compiled with optimized Compiler setttings for the DSP core and some changes to bypass OS specific stuff and to resolve memory alignment issues.

    For you to optimize the floating point code for fixed point processor, you can use Q point math in your code using IQMAth library or optimize floating point math operations by leverage TI`s C64x FastRTS library as a starting point.

    Regards,

    Rahul

    Hi Rahul,

    We have development kits for DM3730 (fixed point DSP) and DM8168 (floating point DSP).

    I have sent an office email to you through "c6swlib_dev@list.ti.com" on 29th Feb 2012 to request for the followings:

    1) The FTP for the C64x+version of OpenCV 1.x.

    2) The FTP for the OpenCV 2.3 on the C674X floating point DSP.

    3) The FTP for the OpenCV 1.x on the C674X floating point DSP.

    Please update.
    Thanks,
    Mr Tan
  • Hi Rahul,

      What is a password for the ftp site? You can reply to Vladimir.Shcherbina@elbitsystems.com an/or to shcherb@gmail.com.

    Best Regards,

      Vladimir

  • Hi everyone,

    I'm looking for OpenCV acceleration using ARM + C674x floating point DSP, it will run on OMAP-L138.

    I'm very interested in this prerelease opencv 2.3 port discussed below: how to achieve it?

    I have also an Igep board with OMAP3530, then fixedpoint C64x, is there a port fot this?

    I would want tu run a sample code to benchmark these two platform in OpenCV application: I 've already code written for x86 and  i want the less effort to reuse it.

    Which is the right and convenient way to interoperate ARM (running some Linux ) and DSP ? C6Accel?

    Is there a better forum section to discuss about my application?

    Thank you!!

  • Hi Rahul,

    My email address:abj578@163.com.

    Could you sned me the FTP link about opencv1.1 c64+ version?

    Many thanks in advance,

    Best Regards, Wangpeng

     

  • Can TI provide the source files or C6accel library of the ported OpenCV function cvCalcOpticalFlowPyrLK ?

    I just need this OpenCV function to run on TI fixed point DSP.

  • I apologize for not responding on this post. We have de-emphasized the support for OpenCV and are no longer distributing the base port until we have a support strategy in place.The only way to get access to the base port of OpenCV on C6000 devices is to contact your local TI representative.

    Regards,

    Rahul

  • Hi Rahul,

    Our local TI representative said that OpenCV port for fixed point DSP cannot be released to us.

    I have no choice but try to port "cvCalcOpticalFlowPyrLK" using C6accel.

    I am using "c6accel_1_01_00_06".

    I try to follow the example in "c6accel_2_01_00_11".

    There is no make error.

    When I run the test application, "cvCalcOpticalFlowPyrLK" finishes in 0.7msec.

    When I do printf on the array status in "cvCalcOpticalFlowPyrLK", it shows all zeroes.

    It means C6accel version of "cvCalcOpticalFlowPyrLK" is called but it is not running properly.

    Please give some advice on how to resolve this problem.

    Can you send me an email so that I can send my source files for you to take a look ?

    I have been with this problem for a week and at loss what to do.

    Attached is the CE log file.

    Please help me.

    Thanks,

    Mr Tan

    1565.CE_log.txt

  • Hi Rahul.

    Please reply to my C6accel issue .

    Thanks,

    Mr Tan

  • Rahul,

    What does 'de-emphasized support' mean?  Also, you then mention that you will 'no longer distribute the base port until you have a strategy in place'.  If I put those 2 items together, are you not saying that 'we're not distributing the base port until further notice'?

    Leave it to a TI person to use 2-3x more words than necessary to say a simple thing. 

    This is why I'm inclined to not use TI products...there's no cohesive approach to open source and startup project support.  It's 1 step forward...1 step sideways...turn yourself around untill dizzy.

    Has TI noticed the buzz around Raspberry Pi?

    Just release the software and sell the hardware...is that a simple enough business model for you?

  • Hi Rahul:

    My e-mail address is simonchu47@gmail.com.

    Could you please send me how to get the OpenCV 2.3 on C674x floating point DSP.

    I am now developing application on DM8168.

    Thank you very much

    Best regards,

    Simon

  • Simon,

    This is no longer distributed on individual basis through an ftp. Please get in touch with your local TI contact to request the software.

    Regards,
    Rahul

  • Hello Rahul,

    I am also in need of things mentioned by Mr. Tan.

    Could you please send FTP links to umeshraddi.kalhal@delphi.com?

    Thanks

    Umesh

  • Hello !!

    Could you please sent me the openCV libriaires for lcdk c6748. the some precompiled in the face detect demo. openCV 1.1.

    My mail is : anouini@gmail.com

    Think you !!