TDA2EVM5777: Infinite values observed in the co-ordinates value

Vyom Mishra1

Genius 4590 points

Part Number: TDA2EVM5777

Dear Sir,

I have Imported the Tensorflow model trained on RGB data using patch released to us "REL.TIDLSRC.01.01.04.00"

Post-processing is not part of the model, it is handled separately.

Please find the output log after post-processing,

Fullscreen 0160.log.txt Download

Output big shape: (1, 24, 19, 19)
Output medium shape: (1, 24, 38, 38)
Output small shape: (1, 24, 76, 76)
Output big shape: (1, 19, 19, 24)
Output medium shape: (1, 38, 38, 24)
Output small shape: (1, 76, 76, 24)
[<tf.Tensor 'Placeholder:0' shape=(?, 19, 19, 24) dtype=float32>, <tf.Tensor 'Placeholder_1:0' shape=(?, 38, 38, 24) dtype=float32>, <tf.Tensor 'Placeholder_2:0' shape=(?, 76, 76, 24) dtype=float32>]
[[      -inf    0.             inf    0.     ]
 [      -inf   67.36842        inf   67.36842]
 [      -inf  202.10527        inf  202.10527]
 [      -inf  471.57898        inf  471.57898]
 [      -inf 1010.52637        inf 1010.52637]
 [      -inf  808.4211         inf  808.4211 ]
 [      -inf  404.21054        inf  404.21054]
 [      -inf  875.7895         inf  875.7895 ]
 [      -inf  538.9474         inf  538.9474 ]
 [      -inf 1145.2632         inf 1145.2632 ]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf    0.             inf    0.     ]
 [      -inf   67.36842        inf   67.36842]
 [      -inf  202.10527        inf  202.10527]
 [      -inf  471.57898        inf  471.57898]
 [      -inf 1010.52637        inf 1010.52637]
 [      -inf  808.4211         inf  808.4211 ]
 [      -inf  404.21054        inf  404.21054]
 [      -inf  875.7895         inf  875.7895 ]
 [      -inf  538.9474         inf  538.9474 ]
 [      -inf 1145.2632         inf 1145.2632 ]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]
 [      -inf       -inf        inf        inf]]
[1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1.]
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
 2 2 2 2 2]

The first list is boxes with xmin ,ymin, xmax, ymax
The second list is the scores for those boxes
The third list is the classes of those boxes

I am finding infinite values for the co-ordinates.

Please find the tensor below:

I have attached import config file for your reference

Fullscreen TF.txt Download

# Default - 0
randParams         = 0 

# 0: Caffe, 1: TensorFlow, Default - 0
modelType          = 1

# 0: Fixed quantization By tarininng Framework, 1: Dyanamic quantization by TIDL, Default - 1
quantizationStyle  = 1 

# quantRoundAdd/100 will be added while rounding to integer, Default - 50
quantRoundAdd      = 50

numParamBits       = 12
# 0 : 8bit Unsigned, 1 : 8bit Signed Default - 1
inElementType      = 0
inQuantFactor      = 255
inputNetFile       = "E:\TIDL_import_tool_v1.2.2\REL.TIDLSRC.01.01.04.00\modules\ti_dl\test\testvecs\config\tidl_models\optimized_DASNet.pb"
inputParamsFile    = "NA"
outputNetFile      = "E:\TIDL_import_tool_v1.2.2\REL.TIDLSRC.01.01.04.00\modules\ti_dl\test\testvecs\config\tidl_models\DASNet_net.bin"
outputParamsFile   = "E:\TIDL_import_tool_v1.2.2\REL.TIDLSRC.01.01.04.00\modules\ti_dl\test\testvecs\config\tidl_models\DASNet_param.bin"

#preProcType  = 4
sampleInData = "E:\TIDL_import_tool_v1.2.2\REL.TIDLSRC.01.01.04.00\modules\ti_dl\test\testvecs\input\PD_RGB.raw"
tidlStatsTool = "E:\TIDL_import_tool_v1.2.2\REL.TIDLSRC.01.01.04.00\modules\ti_dl\utils\quantStatsTool\eve_test_dl_algo.out.exe"

rawSampleInData = 1
inWidth  = 608
inHeight = 608 
inNumChannels = 3

What could be the reason for these infinite value?

Kindly help me to resolve the same.

Thanks and Regards,

Vyom Mishra

over 5 years ago

0 Praveen Eppa1 over 5 years ago

TI__Genius 17580 points

Hi Vyom,

Looks like the scale value applied is not proper.

Did you check the output of last layer (Detection output layer) before post processing?

Thanks,

Praveen

0 Vyom Mishra1 over 5 years ago in reply to Praveen Eppa1

Genius 4590 points

Dear Sir,

"Looks like the scale value applied is not proper"

I am not clear about the above statement, kindly elaborate. if possible kindly suggest the correct one.

We are checking and comparing with PC output of the last layer.

Thanks and Regards,

Vyom Mishra

0 Praveen Eppa1 over 5 years ago in reply to Vyom Mishra1

TI__Genius 17580 points

Hi Vyom

>> I am not clear about the above statement, kindly elaborate. if possible kindly suggest the correct one.

As you said the output log is after the post processing we thought you may not be using proper scale value for the output.

If you are checking and comparing the last layer output then it does not matter, can you share the import tool output log?

Thanks,

Praveen

0 Vyom Mishra1 over 5 years ago in reply to Praveen Eppa1

Genius 4590 points

Dear Sir,

Kindly find the attached import log fro your reference.

3821.log.txt

Thanks and Regards,

Vyom Mishra

0 Vyom Mishra1 over 5 years ago in reply to Vyom Mishra1

Genius 4590 points

Dear Sir,

Gentle Reminder!!

Thanks and Regards,

Vyom Mishra

0 Praveen Eppa1 over 5 years ago in reply to Vyom Mishra1

TI__Genius 17580 points

Hi Vyom,

From the log we can see that final conv layers are having different "out Q"values, so you need to apply scaling on them separately as par their "outQ" values in the post processing.

Please refer to FAQ 10 in the TIDL user guide for more details about out Q.

Thanks,

Praveen

0 Vyom Mishra1 over 5 years ago in reply to Praveen Eppa1

Genius 4590 points

Dear Sir,

As we know we have three outputs( small, medium and big).

So we have provided the outQ values to them separately.

But still, we are facing the infinite values for co-ordinates.

Kindly help us to resolve.

Thanks and Regards,

Vyom Mishra

0 Praveen Eppa1 over 5 years ago in reply to Vyom Mishra1

TI__Genius 17580 points

Hi Vyom,

>> So we have provided the outQ values to them separately.

If so, we expect the output to be correct, but you are seeing some infinite values, so not sure where the problem started.

I would suggest to check the layer level outputs from import tool against the tensorflow reference output, please refer to section 3.8 (Matching TIDL inference result) in the TIDL user guide on how to do this.

Thanks,

Praveen

0 Praveen Eppa1 over 5 years ago in reply to Praveen Eppa1

TI__Genius 17580 points

Hi Vyom,

I haven't heard back from you, I'm assuming you were able to resolve your issue.

If not, just post a reply below (or create a new thread if the thread has locked due to time-out).

Thanks,

Praveen

Processors

Processors forum

TDA2EVM5777: Infinite values observed in the co-ordinates value