AI models off-line

Software
,
1

Hi,

I wanted to use the object_detection.py program given in your Tutorial page. I’m using Python 3.10 on Windows, off-line.

Then when executed, it says that:

  • AI model not found, downloading…
  • File downloading failed
  • Model download error
  • Object detection not found, if connected to the internet it will be downloaded automatically if missing
  • Corrupted SDK installation

I’ve tried to use the Optimize thing on the ZED Diagnostic but it result in “Diagnostic not answering”. So, I’m guessing that it is due to the off-line but I can’t connect my PC to the internet. Do you have a solution to get those AI models to put it in my computer?

2

Hi @uol,

Welcome to the Stereolabs forums :wave:

The ZED SDK indeed requires an internet connection to download the AI models, this is done only once the first time you try to use the model.

What you can do is download the AI models using a computer with an internet connection, and copy the files from the following folder: C:\ProgramData\Stereolabs\resources to your computer without internet access.

3

Hi,

Thank you for your answer.

Does that mean that I have to install the ZED SDK and run it on a computer with internet connection in order to have those files ? Is there an other way to get those models, without downloading the whole thing on another computer?

4

Hi,

Here’s an (unofficial) Python script that will download the ZED SDK 4.1.0 models.
https://gist.githubusercontent.com/adujardin/9f7d82edc6e0246c87b971ce4b009aae/raw/34fc56379bca7ff56b004412180eb308d8bb87dc/download_zed_models.py

This is not guaranteed to have the exact models needed as some are platform-dependent, but should help.

5

Hi ! It’s been a while but I’ve tried your solution.

So, I’ve successfully downloaded the following models into the path C:/ProgramData/Seterolabs/resources:

  • neural_depth_3.6
  • objects_accurate_3.2
  • objects_medium_3.2
  • objects_performance_3.2
  • person_head_acurate_2.4
  • person_head-performance_2.4
  • person_reid_1.4
  • skeleton_body18-3.2
  • skeleton_body38_3.5

By using the program object_detection.py for the first time, models with objects_(...) are optimizing them-self without any problems, and it seem to work find since it can detect a person.

But when I tried to use sl.DEPTH_MODE.NEURAL, it tells me that it can not find a model to optimize. Then, I’ve tried to optimize the model by using the ZED Diagnostic, even with the Windows command in this page: Why is my application frozen when using the NEURAL depth mode? After a few minutes, the ZED Diagnostic window doesn’t respond and none of the models can be optimized.

I’ve also tried using the ZED Depth Viewer, but again when I tried to use the Neural mode, after agreeing to optimized it, it shows this error : Failed to optimized model. NEURAL won't be available.