Unable to run custom model in inference server about acap-computer-vision-sdk-examples HOT 6 CLOSED

Hi, could you ssh in the camera and run
journalctl -u larod

it might give us more informations about the issue

from acap-computer-vision-sdk-examples.

@Corallo
JournalCtl only gives the same message as I got before:
Session 63: Could not run job: Could only read 196608 out of 786432 bytes from file descriptor

from acap-computer-vision-sdk-examples.

Im also curious about the chip id in environment files. If I want to run a TFLite model on the armv7hf CPU, which chip id do i choose? Previously I tried chip id 2 when using CPU models. Are these documented anywhere?

from acap-computer-vision-sdk-examples.

Regarding the chip, it is not properly documented yet, we are going to add a note in the readme, thanks for reporting this issue.

In the meanwhile, yes, -c 2 makes the inference run on cpu. If you have an axis camera with an hardware accellerator you can use -c 4 to run on artpec7 tpu, or -c 12 to run on the artpec8 dlpu.

Regarding the issue with your model, as you mentioned you are using a "custom" model, please be aware that our deep learning API doesn't support custom layers, so it is not enough that the model runs on your local machine.

I'm in touch with the responsible team to see if we can provide you with more information to debug your specific issue, I'll give an update when I have news.

from acap-computer-vision-sdk-examples.

Another possiblity, which is more likely considering the error that you get, is that you are providing the wrong input to your model.
Maybe you supllied the input as uint8 while the model wanted a float32?
196608 [uint8] * 4 = 786432 [float32]

from acap-computer-vision-sdk-examples.

That was actually it, I had forgotten to quantize the inputs of one model to int8. Thanks for pointing out a dumb mistake!

from acap-computer-vision-sdk-examples.