Error while profiling model

取消
显示结果 
显示  仅  | 搜索替代 
您的意思是: 

Error while profiling model

2,884 次查看
Hadrien
Contributor I

Hello to all,

I am new to eIQ, and I am trying to profile my own tensorflow model. I can import my model (.tflite or .h5). I checked that the different layers of the CNN were present, but when I launch the profiling, an error appears (see pictures). Unfortunately the error is not very explicit.

error_profiling_eiQ.pngmessage_error_profiling_eiQ.png

I get the same error as I profile the network:
- from a .tflite saved with the tensorflow library
- from a .h5 saved with the tensorflow library
- from a .tflite saved with eIQ from the loaded .h5 

Does anyone have an idea where the error could be coming from?

Thanks in advance,

 

0 项奖励
5 回复数

2,497 次查看
david_piskula
NXP Employee
NXP Employee

Hello @Hadrien, I don't know if the question is still relevant but did you make sure you have the modelrunner application running on your target?

0 项奖励

2,807 次查看
david_piskula
NXP Employee
NXP Employee

Hello Hadrien, can you share your model or is it confidential? In order to provide the most efficient support, I need to reproduce your issue.

Thanks,

David

0 项奖励

2,786 次查看
Hadrien
Contributor I

Hello David, thank you for your interest,

 

Here is a very minimal exemple :

step 1 in python :

input1 = tf.keras.layers.Input(shape=(1,1024,4))

x_conv_sep = tf.keras.layers.SeparableConv2D(4,
                                                                              [1,3],
                                                                              strides=(1, 1),
                                                                              padding='same',
                                                                              data_format="channels_last",
                                                                              dilation_rate=(1,1),
                                                                              depth_multiplier=2,
                                                                              use_bias=False,
                                                                              input_shape =[1,1024, 4],
                                                                              depthwise_initializer='he_normal',
                                                                              pointwise_initializer='truncated_normal',
                                                                              name = "conv_dilatation",
                                                                             )(input1)

model.save("conv2d_sep.h5")

step 2 in EIQ:

import model

Convert to tflite with quantization (int8) (see picture)

Hadrien_0-1670424524187.png

 

import tflite model 

try to profile...

 

It seems to be the same as explain in  the eIQ Toolkit User Guide 1.5.2 (sec. 4.1) but it doesn't work.

Do you see where is the problem ?

 

Thanks in advance,

 

Hadrien

0 项奖励

2,773 次查看
david_piskula
NXP Employee
NXP Employee

Hello Hadrien,

I will report this behavior to our developers and let you know what's wrong once they figure it out.

Thanks,

David

0 项奖励

2,768 次查看
Hadrien
Contributor I

Thank you for your help

 

Hadrien

0 项奖励