The difference is in the definition on the website and in the purchased SDK [ALPR]

[Viktor Bochikalov]

Hello, we have a situation where the vehicle license plate number is not correctly determined.

But at the same time, the same image is identified correctly on the site.

The original image and comparison are presented in the attachment.

The SDK is launched on an operating system under LINUX [Ubuntu 22.04.4 LTS], server license.

The calculation takes place on the CPU [Intel(R) Xeon(R) CPU E5-2640 v4 @ 2.40GHz]

Ternsorflow library downloaded from your repository:

libtensorflow_r1.14_cpu_linux_x86-64.tar.gz

Why we have different result?

[Mamadou DIOP]

Hi,

Thus has been discussed here several times (check the archive).

You will never have the exact same result when using different hardware/software. They return different result at some point but they will have the same accuracy.

If you try 1000 images on 2 differents servers you'll have same accuracy but they will not always return same result. 

This applies to all AI products. 

Only way to have exact same result: same CPU, same GPU, same CUDA version, same OpenVINO version, same Ubuntu version, same Tensorflow version....

Sent from my Galaxy

--
You received this message because you are subscribed to the Google Groups "doubango-ai" group.
To unsubscribe from this group and stop receiving emails from it, send an email to doubango-ai...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/doubango-ai/9d86d5a1-71f9-4c93-a3b3-d9df90481463n%40googlegroups.com.

[Mamadou DIOP]

Additional remark: you *must* always attach full logs when requesting help. Otherwise, we have no idea what's your env (version, os, hardware...). Most importantly, the logs may contain info about the issue. 

To view this discussion on the web visit https://groups.google.com/d/msgid/doubango-ai/1MeTHG-1siLDp44Em-00aTNP%40mrelayeu.kundenserver.de.

[Mamadou DIOP]

Please keep douba...@googlegroups.com in CC

You'll not loose your license if you change Tensorflow version but this has nothing to do with your issue. 

Our server uses Tensorflow 1.14 but the computation is done on GPU using TensorRT while you're using Tensorflow on CPU and OpenVINO which is totally different env.

Your image is pretty clear, it should not cause issue on T/I

If you want help about your issue you must:

- provide full logs

- provide the code you're using if not recognizer sample

- provide the image causing the issue

Sent from my Galaxy

[Viktor Bochikalov]

Hello, I solved the problem with the difference in recognition.

Disabling Openvino solves my problem [--openvino_enabled False]

The following version is installed on the device:

[PLUGIN_OPENVINO INFO]: [7] OpenVINO inference engine number: 2020.3.0-3467-15f2c61a-releases/2020/3, version: 2.1

Can updating the openvino version to the latest help with recognition? Without openvino, recognition speed decreases by 30%

***

Attached is the log file that was used for recognition.
Used by python_recognizer.sh

четверг, 23 мая 2024 г. в 17:58:58 UTC+6, Mamadou DIOP:

[Mamadou DIOP]

No, you haven't solved your problem. You just switched deep learning framework. Later, you'll find an image not working with Tensorflow but working with OpenVINO and you'll switch again. 

As already explained, different frameworks gives different results but they have same accuracy. 

You have an OCR error and OpenVINO is about detection. So, you definitely didn't solve the issue.

Again, try with the recognizer (cpp) sample and check if you got same error. 

Sent from my Galaxy

-------- Original message --------

From: Viktor Bochikalov <invek...@gmail.com>

Date: 5/24/24 11:02 (GMT+01:00)

To: doubango-ai <douba...@googlegroups.com>

--
You received this message because you are subscribed to the Google Groups "doubango-ai" group.
To unsubscribe from this group and stop receiving emails from it, send an email to doubango-ai...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/doubango-ai/b0463985-ab3a-4112-94ee-fe23348ff48bn%40googlegroups.com.