Model sometimes not running ...

Will

Hi, I noticed that sometimes a model seems to fail to run with the preview showing "NN fps: 0". This happened for both OAK-1 and OAK-D in particular with the face-detection-adas-0001 and vehicle-license-plate-detection-barrier-0106 models using the basic command:
python3 depthai_demo.py -dd -cnn <Model Name>
Re-launching that command few times eventually fixes the problem. I compared the logged messages between working and non-working instances, and noticed that the only difference for the non-working instances was always the same message:
"Successfully opened stream previewout with ID #3!”
while the working instances were always showing a different ID (#2 instead of #3). Let me know if you need more info to troubleshoot this issue. Thanks!

Brandon

Hi Will,

Thanks for the report! (And sorry for the trouble.). If you don't mind, could you see if the same behavior occurs with this example?

https://github.com/luxonis/depthai-experiments/blob/master/gen2_examples/08_rgb_mobilenet.py

We have reworked a lot of the neural network handling in Gen2, and I'm curious if the same issue appears here, or if it is already fixed in the Gen2 system.

Thoughts?

Thanks again,
Brandon

Will

Hi Brandon,
No this Gen2 example does not seem to suffer from the issue I reported for Gen1.
By the way it would be nice if you could update Gen2 example 08_rgb_mobilenet.py with the extraction of the label associated with each detection box.
Thanks again for this great work and associated support !
Will

Brandon

Will Thanks for testing/verifying.

And yes, we should definitely do as you say. Thanks for bringing it up.

So actually this would be a good thing to do to satisfy this Github issue:
https://github.com/luxonis/depthai/issues/159

We could kill 2 birds with one stone on that.

Will ping Philip (who started today) about implementing this. Unfortunately his hardware doesn't arrive until Wednesday, so it might be a bit until he gets going on that.

Thanks again,
Brandon