• State TI Thinks Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 100 subscribers
  • Views 602 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

66AK2H14: OpenCV accuracy failure ?

Naoki Kawada
Guru 19890 points
Part Number: 66AK2H14

Hello,

I walked through the following wiki and found OpenCV accuracy failure table:

I tried the same and saw the same errors with ProcSDK Release 03.02.00 on my K2H EVM. I'm not sure if these failures should be regarded as "fatal failures" or not. Do you have any comments on this ? Is the current OpenCV implementation still an engineering drop ? If yes, do you have any plan for GA(more stable version) ?

Best Regards,
Naoki

  • 0
    TI__Guru 65405 points
    Hi Naoki,

    I've forwarded this to the SW experts. Their feedback should be posted here.

    BR
    Tsvetolin Shulev
  • 0 in reply to Cvetolin Shulev-XID
    TI__Genius 12805 points

    From the developers' team:

    Some of below failures are baseline failures (visible on ARM only platform as well).

    TI DSP specific, kernels can be added (just compiling existing GPU biased OCL kernels is not yielding great results), but I am not really the owner of TI OpenCV.

    we are encouraging customers to add new OpenCL kernels on their own into OpenCV.

    E.g. late last year new OpenCV (baseline) version was released (ocv 3.2). No public plans for q new release

    Ran

  • 0 in reply to ran35366
    TI__Guru 50170 points

    Hi, Noaki,

    There  are explanations for the failures in the table of details of the accuracy tests failures in OpenCV wiki page. However, I got less failures when running the tests on K2H EVM using ProcSDK release 3.2.05. There are 3 modules I need to hit control-c to be out of the hung which have been stall for more than 2 hours. I did a grep on FAIL and only 2 failures shows up in the output file.

    [  FAILED  ] Features2d_AGAST.regression (377 ms)
    [  FAILED  ] Features2d_Detector_AGAST.regression (460 ms)
    [  FAILED  ] 2 tests, listed below:
    [  FAILED  ] Features2d_AGAST.regression
    [  FAILED  ] Features2d_Detector_AGAST.regression
     2 FAILED TESTS

     Please see attached output file from my run for details. The OpenCV in ProcSDK is the GA version. Ran in earlier post also mentioned some possible cause of failures.

    opencv_test_log.txt

  • 0 in reply to Rex Chang
    TI__Guru 50170 points
    correct some typos. Sorry for spelling your name wrong, Naoki. The ProcSDK release is 3.2.0.5.