TDA4VE-Q1: model output is abnormal

? ?

Part Number: TDA4VE-Q1

Hello , TI experts ! We have a fast_BEV model , we cut it to two pieces . The input of the second one 's shape is 1*80*100*100 . We use rtos to import the model , because of the input is not picture , so we have to use binary input formal to import the model . But , when wen import the step2 model , the simulation inference result abnormal . So we compared each layer output of both onnx model and bin model . We find the diffenence is big . So how can we sovle this problem ? we use sdk ti-processor-sdk-rtos-j722s-evm-11_00_00_06 . to_ti.zip

3 months ago

0 Chris Tsongas 3 months ago

TI__Genius 16500 points

Hi,

We are on Thanksgiving holiday in the US and will return Dec 1.

Regards,

Chris

0 Christina Kuruvilla 3 months ago in reply to Chris Tsongas

TI__Expert 7060 points

Hi,

Thank you for sending the files. Could you share your inference configuration? I would like to double check this configuration as well. The model inference completes without outputting any errors or warnings correct?

In the meantime, I will run this on my end and see if I am able to see a similar behavior.

Warm regards,

Christina

0 ? ? 3 months ago in reply to Christina Kuruvilla

Prodigy 90 points

Thanks for your reply

My inference configuration , as follows:

inFileFormat = 1
#postProcType = 2
inElementType = 2

netBinFile = "/home/invo/2TB/evaluate/fast_bev/output/pc_bin/entry/base/import_result_step2/fast_bev_step2_int8.bin"
ioConfigFile = "/home/invo/2TB/evaluate/fast_bev/output/pc_bin/entry/base/import_result_step2/fast_bev_step2_int8_1.bin"
outData = "/home/invo/2TB/evaluate/fast_bev/output/pc_bin/entry/base/infer_result_step2/fast_bev_step2_int8_save.bin"
#inData = "/home/invo/2TB/evaluate/fast_bev/data/data_list_step2_infer.txt"
inData = "/home/invo/2TB/evaluate/fast_bev/data/infer_step2_img/infer_step2_img.raw"
numFrames = 100

#totNumClasses = 2
debugTraceLevel = 3
writeTraceLevel = 3

I did not find errors or warnings .

0 Christina Kuruvilla 3 months ago in reply to ? ?

TI__Expert 7060 points

Thanks for the information! I will update tomorrow with any updates from my testing.

Warm regards,

Christina

0 Christina Kuruvilla 3 months ago in reply to Christina Kuruvilla

TI__Expert 7060 points

Hi,

Apologies in the delay. I am still working on it and should give you an update on Monday on my findings.

Warm regards,

Christina

0 ? ? 3 months ago in reply to Christina Kuruvilla

Prodigy 90 points

We are looking forward to the good message from you ! Thank you very much !

0 Christina Kuruvilla 3 months ago in reply to ? ?

TI__Expert 7060 points

Hi Peng,

Could you share the comparison you have between TIDL and ONNX? I would like to double check as I had changed some of your configurations.

Also, I noticed that you are using a different input for the inference but wasn't added to the zip file. I have been doing testing with the input used for the import, but if you would like me to continue with the same inference, please share.

Warm regards,

Christina

0 Christina Kuruvilla 2 months ago in reply to Christina Kuruvilla

TI__Expert 7060 points

Hi Peng,

Are you still needing help with this? If so, can you please share the comparisons you have between TIDL and ONNX?

Warm regards,

Christina

0 Yong Zhang 1 month ago in reply to Christina Kuruvilla

TI__Mastermind 18741 points

Dear Peng.

would you please help provide the status?

thanks a lot!

yong

0 Yong Zhang 1 month ago in reply to Yong Zhang

TI__Mastermind 18741 points

Dear Peng.

we would like to close this ticket in this week. Please submit new ticket if there is still problem.

thanks a lot!

yong

Processors

Processors forum

TDA4VE-Q1: model output is abnormal