Error while running Case Code

VIM3
#1

Hi!
I have a very simple model - only 2 layers.
I’m unable to get it to run : it exits with the following output.
D [setup_node:367]Setup node id[0] uid[0] op[NBG]
D [print_tensor:129]in : id[ 1] shape[ 28, 28, 1 ] fmt[u8 ] qnt[ASM zp=128, scale=0.007812]
D [print_tensor:129]out: id[ 0] shape[ 256, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003922]
D [optimize_node:311]Backward optimize neural network
D [optimize_node:318]Forward optimize neural network
I [compute_node:260]Create vx node
E [compute_node:286]Create node[0] NBG fail
E [vnn_CreateNeuralNetwork:174]CHECK PTR 174
E [main:210]CHECK PTR 210

How can I debug this?
Note that I can run the inceptionv3 example. I have modified the scripts from that to produce this.

Thanks!

#2

@hmathava How do you run you case code ? Compile directly or replace into demo ？

#3

Thanks for replying!
The following is the workflow I took:

  1. Change and run 0_import_model.sh

    #!/bin/bash

    NAME=my_example
    ACUITY_PATH=…/bin/

    convert_caffe=${ACUITY_PATH}convertcaffe
    convert_tf=${ACUITY_PATH}convertensorflow
    convert_tflite=${ACUITY_PATH}convertflite
    convert_darknet=${ACUITY_PATH}convertdarknet
    convert_onnx=${ACUITY_PATH}convertonnx

    $convert_tf
    –tf-pb ./model/tf_model.pb
    –inputs flatten_input
    –input-size-list ‘28,28’
    –outputs dense/Sigmoid
    –net-output ${NAME}.json
    –data-output ${NAME}.data

  2. Add test images and correct responses to data/validation_tf.txt

  3. Change and run 1_quantize_model.sh

    NAME=my_example
    ACUITY_PATH=…/bin/

    tensorzone=${ACUITY_PATH}tensorzonex

    #asymmetric_quantized-u8 dynamic_fixed_point-8 dynamic_fixed_point-16
    $tensorzone
    –action quantization
    –source text
    –source-file ./data/validation_tf.txt
    –channel-mean-value ‘128 128 128 128’
    –model-input ${NAME}.json
    –model-data ${NAME}.data
    –quantized-dtype asymmetric_quantized-u8
    –quantized-rebuild

    $tensorzone
    –action inference
    –source text
    –source-file ./data/validation_tf.txt
    –channel-mean-value ‘128 128 128 128’
    –model-input ${NAME}.json
    –model-data ${NAME}.data
    –dtype quantized

  4. Change and run 2_export_case_code.sh

    #!/bin/bash

    NAME=my_example
    ACUITY_PATH=…/bin/

    export_ovxlib=${ACUITY_PATH}ovxgenerator

    $export_ovxlib
    –model-input ${NAME}.json
    –data-input ${NAME}.data
    –channel-mean-value ‘128 128 128 128’
    –reorder-channel ‘0 1 2’
    –export-dtype quantized
    –model-quantize ${NAME}.quantize
    –optimize VIPNANOQI_PID0X88
    –viv-sdk …/bin/vcmdtools
    –pack-nbg-unify

    rm *.h *.c .project .cproject *.vcxproj *.lib BUILD *.linux

    mv nbg_unify nbg_unify_${NAME}

    cd nbg_unify_${NAME}

    mv network_binary.nb ${NAME}.nb

  5. Copy build_vx.sh and run that. Copy built executable and nb file to vim3 pro board and run.

Please let me know if I missed anything. Thanks!

#4

@hmathava You model is a single channel model . Not a three channel model . So,--channel-mean-value can’t be ‘128 128 128 128’

Screenshot from 2020-04-14 10-25-18
Screenshot from 2020-04-14 10-25-18925×601 208 KB

#5

Thanks for noticing that -I will change it.
What is that document you have attached? Where can I find that?

#6

@hmathava It should be in SDK

#7

Thanks for the inputs.
I have changed the scripts to reflect channel means parameter of 0 0 0 256 but I still get the same error.

#8

Hello again,
Is there anything else you’d like me to change with my current process?

#9

@hmathava Can you provide a complete running log information ?

#10

Here it is:
D [setup_node:367]Setup node id[0] uid[0] op[NBG]
D [print_tensor:129]in : id[ 1] shape[ 28, 28, 1 ] fmt[u8 ] qnt[ASM zp= 0, scale=0.003906]
D [print_tensor:129]out: id[ 0] shape[ 10, 1 ] fmt[f16] qnt[NONE]
D [optimize_node:311]Backward optimize neural network
D [optimize_node:318]Forward optimize neural network
I [compute_node:260]Create vx node
E [compute_node:286]Create node[0] NBG fail
E [vnn_CreateNeuralNetwork:174]CHECK PTR 174
E [main:210]CHECK PTR 210

#11

@hmathava This is the first time I have met this situation. The value prompt nbg create fail has no specific error type prompt and there is no way to judge .

#12

Thats unfortunate :frowning:
I am trying to run a single dense layer through you API to gauge the performance. What would be your suggested methodology?

#13

@hmathava Maybe you can tell me that what is the purpose of your model and what are the inputs and outputs?

#14

My model is to just test the TOPs of the system using Dense Layers.
Currently it is:
(28,28)Input -> Flatten -> Dense(256 w/ Sigmoid) -> Softmax

#15

@hmathava Do you want to evaluate the performance of those NPU ?

#16

Yes I would like to evaluate performance specifically for Dense layers.

#17

@hmathava Maybe you can use a google model to test . I find google model is always a good conversion