NN models

[hwaccel]

So I may of missed it in the documentation, but in each of the examples with nn (yolov3,yolov4, ssd, moblienet) reference the blobs located in examples\models but there is no such folder or blob files. Where can i find the premade blob files for the oak-d camera?

From 08_rgb_moblienet line 10 nnPathDefault = str((Path(file).parent / Path('models/mobilenet-ssd_openvino_2021.2_6shave.blob')).resolve().absolute())

Thanks,

[cafemoloko]

@hwaccel link to the blob is here. Please create the folder and copy the blob. Also linked in the documentation page

[hwaccel]

oh thank you for the reply I must of just skimmed over that section, is their a reason to not have it included or complied on demand like the in the original depthai code?

[cafemoloko]

@hwaccel I'll bring this up with the team. Maybe this information (that you need to download the blob) should be more visible. Thank you for the feedback!

[hwaccel]

Yep I'm still very much so learning the oak-d system and this change to gen2 pipeline so I was expecting it to be the same as the depthai_demo.py script (from gen1) where you put in the nn option and it will build the blob for you and you can modify the setting for shave so you can learn about those options.

[i008128]

Hi I would recommend create a step to automatically update all the requested libraries to run the ML models. It is not make any sense to do it manually

[VanDavv]

@i008128 @hwaccel these examples were made to show how to use Gen2 API on a more low level perspective. Since this repository contains whole OAK Python API, we don't want to grow it in size by storing the blobs here.

We're working now on updating depthai_demo.py script, but in the mean time, if you want more examples that work out of the box, please take a look at depthai-experiments repository, where we keep a complete use-case oriented examples (whereas the examples in this repository are API-oriented)

[Luxonis-Brandon]

And totally agreed @i008128 . So we are working to make all of these just as easy. Here is the plan overall:

https://github.com/luxonis/depthai-docs-website/issues/163

Any thoughts/comments?

Anyway, the idea is to start with examples that just work, where no user has to download models, move them in to a directory/etc. These should all just work.

Thoughts?

[SzabolcsGergely]

Model downloader for blobs was merged into both main and develop.