UltimateALPR v4.0 in progress, share your issues

[Mamadou DIOP]

Hello,

We're working on UltimateALPR v4.0 and planning to replace Tensorflow
with Pytorch and use different architectures to solve some well-known
issues.

I'm pinning this post so that you can comment it with the issues you're
facing in v3.

Datasets are welcome.

Regards,

[doubango-ai]

I wanted to address some persistent requests done in this group by providing a patch hosted at  https://doubango.org/SDKs/anpr/ualpr_v3_15.patch

These are low priority requests for us and will be address in v4.

The patch adds:

- supports for Tensorflow 2.18.0 (CUDA 12.5), Tensorflow 2.16.1 (CUDA 12.3), Tensorflow 2.14.0 (CUDA 11.8)

- supports for TensorRT 10.x (CUDA 12.x) on Windows and Linux x86_64

- supports for Jetpack 6.2.0 and 7.0.0 on Jetson
this is not an official release, use it carefully (not fully vetted)

[Jack Heaffey]

Hi,

My issues with UltimateALPR are the following:

-  No support for the i.MX 8M Plus NPU

-  The VMMR models have not been updated in 3 years despite the documentation stating that they would be updated every 6 months -  https://www.doubango.org/SDKs/anpr/docs/Features.html#features-vehiclemakemodelrecognition

-  No support for Arabic or Cyrillic character sets

Jack

[MARIO SHIN]

Hello.

This is an issue I've raised before: Korean license plates frequently contain errors where the character ‘배’ is incorrectly recognized as ‘바’.
In that thread, you suggested a solution to convert ‘바’ to ‘배’ in the code. However, since license plates containing the ‘바’ character already exist, this approach does not resolve the problem.

The images attached below are examples of this error occurring in the test on the https://www.doubango.org/webapps/alpr/ page.

I sincerely hope this is improved in version v4.0.
Thank you.

2025년 10월 31일 금요일 오전 6시 55분 17초 UTC+9에 Mamadou DIOP님이 작성:

[Mamadou DIOP]

Hi,

On 11/13/2025 5:36 PM, Jack Heaffey wrote:

Hi,

My issues with UltimateALPR are the following:

-  No support for the i.MX 8M Plus NPU

It's on the priority list

-  The VMMR models have not been updated in 3 years despite the documentation stating that they would be updated every 6 months -  https://www.doubango.org/SDKs/anpr/docs/Features.html#features-vehiclemakemodelrecognition

Is done

-  No support for Arabic or Cyrillic character sets

We don't have large enough Arabic dataset yet. I don't know of any country with Cyrillic chars in its plates

Jack

On Thursday, 30 October 2025 at 21:55:17 UTC Mamadou DIOP wrote:

Hello,

We're working on UltimateALPR v4.0 and planning to replace Tensorflow
with Pytorch and use different architectures to solve some well-known
issues.

I'm pinning this post so that you can comment it with the issues you're
facing in v3.

Datasets are welcome.

Regards,

--
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 visit https://groups.google.com/d/msgid/doubango-ai/3c0b0fcf-f5d8-473d-a78b-7df808e06893n%40googlegroups.com.

[VISHNU P]

when will the v4 of alpr release like the approximate production time?

To view this discussion visit https://groups.google.com/d/msgid/doubango-ai/e9ddefc7-4e55-49b8-a339-8bd8dc612f49%40doubango.org.

[Mamadou DIOP]

ETA 2 months from now if everything is OK. Last time I said 2 months for a product (https://groups.google.com/g/doubango-ai/c/7q_Uo5Tfr_I/m/-fYMHjcJFAAJ), it took 4 months.

All pytorch models (OCR, VMMR, LPCI...) are done. We're working on the NPU backends (i.MX, Khadas, OrangePi...) and not everything is OK :-