• State Resolved
  • Locked Locked
  • Replies 8 replies
  • Answers 1 answer
  • Subscribers 98 subscribers
  • Views 874 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.

TIDL SSD based object detection - partitioning into EVE & DSP

Koichi Saito
Intellectual 2525 points

Other Parts Discussed in Thread: TDA2

Hi,

In TIDeepLeraningLibrary_DataSheet.pdf (TIDL 01.01.00.00), folloiwng comment is shown in TABLE.4B on page 4.

 " Initial part of the network is executed / profiled on EVE and the later part is on C6x DSP (Later part of the network is best optimal on C6x DSP). "

SSD based detection network in Caffe-Jacinto-Model contains 45 layers as below.
   0, TIDL_DataLayer
   1, TIDL_BatchNormLayer
   2, TIDL_ConvolutionLayer
   3, TIDL_ConvolutionLayer
   4, TIDL_ConvolutionLayer
   5, TIDL_ConvolutionLayer
   6, TIDL_ConvolutionLayer
   7, TIDL_ConvolutionLayer
   8, TIDL_PoolingLayer
     ...
   40, TIDL_FlattenLayer
   41, TIDL_ConcatLayer
   42, TIDL_ConcatLayer
   43, TIDL_DetectionOutputLayer
   44, TIDL_DataLayer

[Q1] To get both EVE and DSP performance shown in TABLE.4B, how do we partition into EVE and DSP ? For instance, which layers are executed on DSP and which other layers are executed on EVE ?

[Q2] To get both EVE and DSP performance shown in TABLE.4B, how do we test on target EVM ?  I need to duplicate the test.

Best regards,
Saito

  • 0
    TI__Genius 17580 points
    Hi Saito,

    I have explained that in below e2e thread, please refer to my last two responses in this thread,

    e2e.ti.com/.../690888

    Thanks,
    Praveen
  • 0 in reply to Praveen Eppa1
    TI__Intellectual 2525 points

    Thanks Praveen.

    On Q2, should we run TIDL OD usecase in VSDK ?   If so, the TIDL OD usecase can run on TDA2 SoC only ?

    Best regards,
    Saito

  • 0 in reply to Koichi Saito
    TI__Genius 17580 points
    Hi Satio,

    You can get the performance numbers by running .out files on standalone, refer to section 3.3.5 (Building the Test Application Executable through GMAKE) to know the steps.

    Thanks,
    Praveen
  • 0 in reply to Praveen Eppa1
    TI__Intellectual 2525 points

    Hi Praveen,

      I have already run .out file on EVE standalone. However I did not get the performance numbers described in TABLE 4B TIDeepLeraningLibrary_DataSheet.pdf.

      Please find attached CCS log.

    Fullscreen LOG_TDA2Px_EVE_jdetnet.txt Download 
    Processing config file C:\ti\psdk\PSDKV33\ti_components\algorithms\REL.TIDL.01.01.00.00\modules\ti_dl\test\testvecs\config\infer\tidl_config_jdetnet.txt !
  0, TIDL_DataLayer                ,  0,  -1 ,  1 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  0 ,    0 ,    0 ,    0 ,    0 ,    1 ,    3 ,  320 ,  768 ,
  1, TIDL_BatchNormLayer           ,  1,   1 ,  1 ,  0 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  1 ,    1 ,    3 ,  320 ,  768 ,    1 ,    3 ,  320 ,  768 ,
  2, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  1 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  2 ,    1 ,    3 ,  320 ,  768 ,    1 ,   32 ,  160 ,  384 ,
  3, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  2 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  3 ,    1 ,   32 ,  160 ,  384 ,    1 ,   32 ,   80 ,  192 ,
  4, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  3 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  4 ,    1 ,   32 ,   80 ,  192 ,    1 ,   64 ,   80 ,  192 ,
  5, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  4 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  5 ,    1 ,   64 ,   80 ,  192 ,    1 ,   64 ,   40 ,   96 ,
  6, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  5 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  6 ,    1 ,   64 ,   40 ,   96 ,    1 ,  128 ,   40 ,   96 ,
  7, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  6 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  7 ,    1 ,  128 ,   40 ,   96 ,    1 ,  128 ,   40 ,   96 ,
  8, TIDL_PoolingLayer             ,  1,   1 ,  1 ,  7 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  8 ,    1 ,  128 ,   40 ,   96 ,    1 ,  128 ,   20 ,   48 ,
  9, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  8 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  9 ,    1 ,  128 ,   20 ,   48 ,    1 ,  256 ,   20 ,   48 ,
 10, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  9 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 10 ,    1 ,  256 ,   20 ,   48 ,    1 ,  256 ,   10 ,   24 ,
 11, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 10 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 11 ,    1 ,  256 ,   10 ,   24 ,    1 ,  512 ,   10 ,   24 ,
 12, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 11 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 12 ,    1 ,  512 ,   10 ,   24 ,    1 ,  512 ,   10 ,   24 ,
 13, TIDL_PoolingLayer             ,  1,   1 ,  1 , 12 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 13 ,    1 ,  512 ,   10 ,   24 ,    1 ,  512 ,    5 ,   12 ,
 14, TIDL_PoolingLayer             ,  1,   1 ,  1 , 13 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 14 ,    1 ,  512 ,    5 ,   12 ,    1 ,  512 ,    3 ,    6 ,
 15, TIDL_PoolingLayer             ,  1,   1 ,  1 , 14 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 15 ,    1 ,  512 ,    3 ,    6 ,    1 ,  512 ,    2 ,    3 ,
 16, TIDL_ConvolutionLayer         ,  1,   1 ,  1 ,  7 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 16 ,    1 ,  128 ,   40 ,   96 ,    1 ,  256 ,   40 ,   96 ,
 17, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 12 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 17 ,    1 ,  512 ,   10 ,   24 ,    1 ,  256 ,   10 ,   24 ,
 18, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 13 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 18 ,    1 ,  512 ,    5 ,   12 ,    1 ,  256 ,    5 ,   12 ,
 19, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 14 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 19 ,    1 ,  512 ,    3 ,    6 ,    1 ,  256 ,    3 ,    6 ,
 20, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 15 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 20 ,    1 ,  512 ,    2 ,    3 ,    1 ,  256 ,    2 ,    3 ,
 21, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 16 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 21 ,    1 ,  256 ,   40 ,   96 ,    1 ,   16 ,   40 ,   96 ,
 22, TIDL_FlattenLayer             ,  1,   1 ,  1 , 21 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 22 ,    1 ,   16 ,   40 ,   96 ,    1 ,    1 ,    1 ,61440 ,
 23, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 16 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 23 ,    1 ,  256 ,   40 ,   96 ,    1 ,   16 ,   40 ,   96 ,
 24, TIDL_FlattenLayer             ,  1,   1 ,  1 , 23 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 24 ,    1 ,   16 ,   40 ,   96 ,    1 ,    1 ,    1 ,61440 ,
 25, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 17 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 26 ,    1 ,  256 ,   10 ,   24 ,    1 ,   24 ,   10 ,   24 ,
 26, TIDL_FlattenLayer             ,  1,   1 ,  1 , 26 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 27 ,    1 ,   24 ,   10 ,   24 ,    1 ,    1 ,    1 , 5760 ,
 27, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 17 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 28 ,    1 ,  256 ,   10 ,   24 ,    1 ,   24 ,   10 ,   24 ,
 28, TIDL_FlattenLayer             ,  1,   1 ,  1 , 28 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 29 ,    1 ,   24 ,   10 ,   24 ,    1 ,    1 ,    1 , 5760 ,
 29, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 18 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 31 ,    1 ,  256 ,    5 ,   12 ,    1 ,   24 ,    5 ,   12 ,
 30, TIDL_FlattenLayer             ,  1,   1 ,  1 , 31 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 32 ,    1 ,   24 ,    5 ,   12 ,    1 ,    1 ,    1 , 1440 ,
 31, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 18 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 33 ,    1 ,  256 ,    5 ,   12 ,    1 ,   24 ,    5 ,   12 ,
 32, TIDL_FlattenLayer             ,  1,   1 ,  1 , 33 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 34 ,    1 ,   24 ,    5 ,   12 ,    1 ,    1 ,    1 , 1440 ,
 33, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 19 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 36 ,    1 ,  256 ,    3 ,    6 ,    1 ,   24 ,    3 ,    6 ,
 34, TIDL_FlattenLayer             ,  1,   1 ,  1 , 36 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 37 ,    1 ,   24 ,    3 ,    6 ,    1 ,    1 ,    1 ,  432 ,
 35, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 19 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 38 ,    1 ,  256 ,    3 ,    6 ,    1 ,   24 ,    3 ,    6 ,
 36, TIDL_FlattenLayer             ,  1,   1 ,  1 , 38 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 39 ,    1 ,   24 ,    3 ,    6 ,    1 ,    1 ,    1 ,  432 ,
 37, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 20 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 41 ,    1 ,  256 ,    2 ,    3 ,    1 ,   16 ,    2 ,    3 ,
 38, TIDL_FlattenLayer             ,  1,   1 ,  1 , 41 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 42 ,    1 ,   16 ,    2 ,    3 ,    1 ,    1 ,    1 ,   96 ,
 39, TIDL_ConvolutionLayer         ,  1,   1 ,  1 , 20 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 43 ,    1 ,  256 ,    2 ,    3 ,    1 ,   16 ,    2 ,    3 ,
 40, TIDL_FlattenLayer             ,  1,   1 ,  1 , 43 ,  x ,  x ,  x ,  x ,  x ,  x ,  x , 44 ,    1 ,   16 ,    2 ,    3 ,    1 ,    1 ,    1 ,   96 ,
 41, TIDL_ConcatLayer              ,  1,   5 ,  1 , 22 , 27 , 32 , 37 , 42 ,  x ,  x ,  x , 46 ,    1 ,    1 ,    1 ,61440 ,    1 ,    1 ,    1 ,69168 ,
 42, TIDL_ConcatLayer              ,  1,   5 ,  1 , 24 , 29 , 34 , 39 , 44 ,  x ,  x ,  x , 47 ,    1 ,    1 ,    1 ,61440 ,    1 ,    1 ,    1 ,69168 ,
 43, TIDL_DetectionOutputLayer     ,  1,   2 ,  1 , 46 , 47 ,  x ,  x ,  x ,  x ,  x ,  x , 48 ,    1 ,    1 ,    1 ,69168 ,    1 ,    1 ,    1 ,  560 ,
 44, TIDL_DataLayer                ,  0,   1 , -1 , 48 ,  x ,  x ,  x ,  x ,  x ,  x ,  x ,  0 ,    1 ,    1 ,    1 ,  560 ,    0 ,    0 ,    0 ,    0 ,
TEST_REPORT_INIT_PROFILE_DATA : TSC cycles = 1125769866, SCTM VCOP BUSY cycles = 0, SCTM VCOP Overhead = 0
I/O Bytes :        0 (        0 +        0 ) :        0 (        0 +        0 ) 

Processing Frame Number : 0 

TEST_REPORT_PROCESS_PROFILE_DATA : TSC cycles = 1607498494, SCTM VCOP BUSY cycles = 70089004, SCTM VCOP Overhead = 0
I/O Bytes :        0 (        0 +        0 ) :        0 (        0 +        0 )


      [Q2] How do we get the performance numbers in TALBE 4B ?

    Best regards,
    Saito

  • 0 in reply to Koichi Saito
    TI__Genius 17580 points
    Hi Satio,

    Can you share infer config file used for running the .out file?

    Thanks,
    Praveen
  • 0 in reply to Praveen Eppa1
    TI__Intellectual 2525 points

    Hi Praveen,

    I used "tidl_config_jdetnet.txt" in directory REL.TIDL.01.01.00.00\modules\ti_dl\test\testvecs\config\infer.

    Best regards,
    Saito

  • 0 in reply to Koichi Saito
    TI__Genius 17580 points

    Hi Satio,

    This "tidl_config_jdetnet.txt" file runs full Net on EVE, but we need to part of network in EVE and later part on DSP to get the performance numbers, so please use attached config files

    First run "tidl_config_jdetnet_eve.txt" on EVE which gives performance numbers of EVE

    Then run "tidl_config_jdetnet_dsp.txt" on DSP which gives performance numbers of DSP

    Thanks,

    Praveen

    Fullscreen 5444.tidl_config_jdetnet_eve.txt Download 
    rawImage    = 1
numFrames   = 1
preProcType  = 4
#inData   = "D:\work\vision\dsp_git\dsp_apps\modules\ti_dl\test\testvecs\config\caffe_models\SSD\000535.jpg"
#inData   = D:\work\vision\dsp_git\dsp_apps\modules\ti_dl\test\testvecs\config\caffe_models\SSD\000220.jpg
#inData   = "D:\work\vision\CNN\data\tiscapes\VIRB0003_200frames_00.11\out000018_crop.png"
#inData   = "D:\work\vision\CNN\data\tiscapes\V105Crop_200frames_0249\out000026.png"
traceDumpBaseName   = ".\trace_dump_"
updateNetWithStats   = 0
netBinFile     = ..\..\test\testvecs\config\tidl_models\jdetnet\tidl_net_jdetNet_ssd.bin
paramsBinFile        = ..\..\test\testvecs\config\tidl_models\jdetnet\tidl_param_jdetNet_ssd.bin

readQ  = 1
writeQ = 1
#inData   = "..\..\test\testvecs\input\trace_dump_0_768x384.y"
inData   = "..\..\test\testvecs\input\trace_dump_0_768x320.y"
outData   = ".\stats_tool_out_eve.bin"
#inData   = ".\stats_tool_out_eve.bin"
#outData   = ".\stats_tool_out.bin"
layersGroupId = 1
runFullNet = 0

    Fullscreen 6332.tidl_config_jdetnet_dsp.txt Download 
    rawImage    = 1
numFrames   = 1
preProcType  = 4
#inData   = "D:\work\vision\dsp_git\dsp_apps\modules\ti_dl\test\testvecs\config\caffe_models\SSD\000535.jpg"
#inData   = D:\work\vision\dsp_git\dsp_apps\modules\ti_dl\test\testvecs\config\caffe_models\SSD\000220.jpg
#inData   = "D:\work\vision\CNN\data\tiscapes\VIRB0003_200frames_00.11\out000018_crop.png"
#inData   = "D:\work\vision\CNN\data\tiscapes\V105Crop_200frames_0249\out000026.png"
traceDumpBaseName   = ".\trace_dump_"
updateNetWithStats   = 0
netBinFile     = ..\..\test\testvecs\config\tidl_models\jdetnet\tidl_net_jdetNet_ssd.bin
paramsBinFile        = ..\..\test\testvecs\config\tidl_models\jdetnet\tidl_param_jdetNet_ssd.bin

readQ  = 1
writeQ = 1
#inData   = "..\..\test\testvecs\input\trace_dump_0_768x384.y"
#inData   = "..\..\test\testvecs\input\trace_dump_0_768x320.y"
#outData   = ".\stats_tool_out_eve.bin"
inData   = ".\stats_tool_out_eve.bin"
outData   = ".\stats_tool_out.bin"
layersGroupId = 2
runFullNet = 0

  • 0 in reply to Praveen Eppa1
    TI__Intellectual 2525 points
    Hi Praveen,

    My issue resolved. Thank you.

    Best regards,
    Saito