Issues Running Custom Model with object-detector-python Sample on Axis Camera Docker

[Hyenni]

Hello,

I'm encountering difficulties while attempting to test my custom developed model using the object-detector-python sample provided in the Axis Communications GitHub repository. Below are the steps I've taken and the corresponding issue I've encountered:

1. I successfully executed the object-detector-python sample code with the default setup.
   

2. I've installed Docker, set up TLS, and configured the ACAP runtime environment using the following commands:

$ export ARCH=aarch64
$ export CHIP=artpec8
$ export APP_NAME=acap4-object-detector-python
$ export MODEL_NAME=acap-dl-models

$ openssl req -new -key server-key.pem -out server.csr -config ./openssl.cnf
$ openssl x509 -req -in server.csr -CA ca.pem -CAkey ca-key.pem -CAcreateserial -out server-cert.pem -days 365 -sha256 -extfile ./openssl.cnf -extensions v3_req
$ scp ca.pem server-cert.pem server-key.pem [email protected]:/usr/local/packages/dockerdwrapper/

$ docker run --rm --privileged multiarch/qemu-user-static --credential yes --persistent yes
$ docker build --tag $APP_NAME --build-arg ARCH .
$ docker build --file Dockerfile.model --tag $MODEL_NAME --build-arg ARCH .
$ DEVICE_IP=X.X.X.X
$ DOCKER_PORT=2376
$ export DOCKER_CERT_PATH=/d/hinkim자료/Docker/acap
$ docker --tlsverify --host tcp://$DEVICE_IP:$DOCKER_PORT system prune --all --force
$ docker save -o app_image.tar $APP_NAME
$ docker --tlsverify --host tcp://$DEVICE_IP:$DOCKER_PORT load -i app_image.tar
$ docker save -o model_image.tar $MODEL_NAME
$ docker --tlsverify --host tcp://$DEVICE_IP:$DOCKER_PORT load -i model_image.tar
$ export ACAP_RUNTIME=axisecp/acap-runtime:1.3.1-aarch64-containerized
$ docker save $ACAP_RUNTIME -o acap_runtime.tar
$ docker --tlsverify --host tcp://$DEVICE_IP:$DOCKER_PORT load -i acap_runtime.tar

$ openssl req -x509 -batch -subj '//CN=X.X.X.X' -days 365 -newkey rsa:4096 -nodes -out server.pem -keyout server.key -nodes -out server.pem -keyout server.key
$ scp server.pem server.key root@X,X,X,X:/usr/local/packages/acapruntime
$ ssh root@X,X,X,X 'chown sdk /usr/local/packages/acapruntime/server.*'

3. I modified the original code by updating it with my model and version details and proceeded with the build process without any issues.
   docker-compose.yml
   
   Dockerfile.model
   

4. Additionally, the model I intend to upload is yolov8_PPE_tflite, and I have confirmed that both the model and the label file have been successfully uploaded and are listed in the model list output in the CMD window.

5. Upon execution, I encountered the following error:
   

6. I found a similar issue in the GitHub repository, which suggested a potential problem with the ACAP runtime version. However, I've already used the curl command as guided and downloaded the indicated version (1.3.1), which I've uploaded as a tar file.
   
   https://github.com/orgs/AxisCommunications/discussions/331

I've followed the guide meticulously, yet I'm still facing this error. Any guidance to resolve this problem would be greatly appreciated.

Thank you for your assistance.

+++ I checked one more related issue in Git, and I checked the comment that it might be a memory issue.
I checked the memory and it's small. So is there a possibility that it's an issue caused by low memory capacity?
For your information, The camera I'm using is P3265-LV

https://github.com/orgs/AxisCommunications/discussions/316#discussioncomment-7598266

[github-actions[bot]]

This automatically generated reply acts as a friendly reminder.

Answers to your questions will most often come from the community, from developers like yourself. You will, from time to time, find that Axis employees answers some of the questions, but this is not a guarantee. Think of the discussion forum as a complement to other support channels, not a replacement to any of them.

If your question remains unanswered for a period of time, please revisit it to see whether it can be improved by following the guidelines listed in Axis support guidelines.

[vivekatoffice]

Hi @Hyenni ,
Please use the following link and share your output.

[Hyenni]

Hello @vivekatoffice

Thank you for your quick reply.
I'm sharing the output, so I'd appreciate it if you could check it.

[vivekatoffice]

Hi @Hyenni ,
I suggest you follow this example from here:
Training and converting a model to Tensorflow Lite format

Look like your model need to be quantized from 32-bit floating-point (FP32) precision to 8-bit integer (INT8) precision. Let's see what @Corallo 's feedback or suggestion 👍

[Hyenni]

Hello @vivekatoffice

Thank you for your reply
I'll try to convert model again according to the guide you sent me.

[vivekatoffice]

Yolov5 on Artpec-8 instructions:

Hi,
It is now possible to run Yolov5 on Artpec-8 on devices running Axis OS >= 11.7
See the guide here

Originally posted by @Corallo in #462 (comment)

[Corallo]

I'd like to add also that your model is probably not compatible with Artpec 8.
Before building the application I would suggest trying the model in isolation.
You can find instructions in the ACAP documentation:
https://axiscommunications.github.io/acap-documentation/docs/computer-vision-on-device/test-your-model.html

[Hyenni]

Hello @Corallo

Thank you for your response.
I checked the output by referring to the link you shared.
I would appreciate it if you could check it out.

[Corallo]

It sounds like that layer mentioned in the log is (one of) the non-compatible layers.
Also, from the name of the model I see it is float32, which is not supported by the DLPU.
There might be many issues preventing you to run Yolov8, if you are unsure I would recommend to rather follow the guide @vivekatoffice linked before. Yolov5 should not give you compatibility issues.

To know more about what are the limitations of running a model on DLPU, check the documentation:
https://axiscommunications.github.io/acap-documentation/docs/computer-vision-on-device

[Hyenni]

Hello @Corallo

Thank you for your reply
I'll try the convert model again as @vivekatoffice sent me the link, and if it doesn't work, I'll refer to the document you shared

[vivekatoffice]

Hi @Hyenni ,
Is there any update or alternative solution available for this topic?

Thanks!