topic Re: Can not get correct inference result in eIQ Machine Learning Software

Can not get correct inference result

lixianghan — Wed, 15 Sep 2021 10:12:21 GMT

I try to follow this tutorial: Getting Started with TensorFlow Lite for Microcontrollers on i.MX RT to deploy flowers classification model in i.MX RT1050-EVKB. I test the generated .tflite model following the tutorial, and get correct inference result. But I can not get correct inference result after converted .tflite to .h with xxd, and deploy it in i.MX RT1050-EVKB.

Re: Can not get correct inference result

david_piskula — Thu, 16 Sep 2021 08:35:31 GMT

Hello @lixianghan,

thank you for your interest in eIQ. Could you please try converting your flower image with the following python script instead of xxd?:

import cv2 import numpy as np img = cv2.imread('flower.bmp') img = cv2.resize(img, (128, 128)) img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) with open('image_data.h', 'w') as fout: print('static const uint8_t image_data[] = {', file=fout) img.tofile(fout, ', ', '0x%02X') print('}\n', file=fout)

Thanks,

David

Re: Can not get correct inference result

lixianghan — Thu, 16 Sep 2021 09:45:33 GMT

Thank you for your help! I solved this problem.

The correct python script:

------------------------------------------------------------------------------------------

import cv2

import numpy as np

img = cv2.imread('daisy.bmp')

img = cv2.resize(img, (128, 128))

img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

with open('daisy.h', 'w') as fout:

print('#define STATIC_IMAGE_NAME "daisy"', file=fout)

print('const char daisy_bmp[] = {', file=fout)

img.tofile(fout, ', ', '0x%02X')

print('};\n', file=fout)

------------------------------------------------------------------------------------------

The ; in the last is important.

Re: Can not get correct inference result

david_piskula — Thu, 16 Sep 2021 09:57:54 GMT

Great! Sorry about the missing semicolon. We will make sure to update the lab guide conversion, thanks for trying it out and letting us know about it.

Have fun with eIQ.

Best Regards,

David