Update base image

[NickM-27]

Proposed change

Updates the base image to debian bullseye

Type of change

• Dependency upgrade
• Bugfix (non-breaking change which fixes an issue)
• New feature
• Breaking change (fix/feature causing existing functionality to break)
• Code quality improvements to existing code
• Documentation Update

Additional information

• This PR fixes or closes issue: fixes #
• This PR is related to issue:

Checklist

• The code change is tested and works locally.
• Local tests pass. Your PR cannot be merged unless tests pass
• There is no commented out code in this PR.
• The code has been formatted using Ruff (ruff format frigate)

[NickM-27]

CC @MarcA711

Looks like there shouldn't be any required changes on rockchip side but just wanted to make sure you are aware of this

[MarcA711]

@NickM-27 I will test it ASAP. Perhaps over the weekend. Thank you for notifying me.

[MarcA711]

Hey @NickM-27, I am currently looking into updating the rockchip image. I want to update the detection models. It seems yolonas from super-gradients was abandoned, at least the URLs for downloading yolonas are broken and there is no recent activity in the repo or support on issues. So this notebook from the frigate repo will also not works: https://github.com/blakeblackshear/frigate/blob/dev/notebooks/YOLO_NAS_Pretrained_Export.ipynb

What model should I use? Are ultralytics models still not useable due to the license issues? Or do newer models like yolo11 have another license?

Moreover, I saw that there are new AI features in the new frigate version. Is there a way to accelerate this on different hardware (similar to detectors)?

[NickM-27]

We can take another look and see if there is a way to get the YOLONAS notebook working.

What model should I use? Are ultralytics models still not useable due to the license issues? Or do newer models like yolo11 have another license?

I am not entirely sure what the current rockchip options are, nothing has changed regarding ultralytics models, I believe there are some re-implementations of some of the YOLO model architectures without the AGPL3 license.

Moreover, I saw that there are new AI features in the new frigate version. Is there a way to accelerate this on different hardware (similar to detectors)?

Many platforms are supported, all of this is done through ONNX runtime. It looks like Rockchip has an ONNX runtime execution provider that is currently in preview state https://onnxruntime.ai/docs/execution-providers/community-maintained/RKNPU-ExecutionProvider.html

[MarcA711]

We can take another look and see if there is a way to get the YOLONAS notebook working.

Someone mentioned here that one can download the weights manually. I will try to do this.

I  am not entirely sure what the current rockchip options are, nothing has changed regarding ultralytics models, I believe there are some re-implementations of some of the YOLO model architectures without the AGPL3 license.

The ones that are supported best and are optimized for Rockchip are listed here. However, almost every object detection model should work.

Many platforms are supported, all of this is done through ONNX runtime. It looks like Rockchip has an ONNX runtime execution provider that is currently in preview state https://onnxruntime.ai/docs/execution-providers/community-maintained/RKNPU-ExecutionProvider.html

Unfortunately this is for the old generation of socs (not supported by Frigate and used by fewer people). Moreover, this has been in Preview state with no progress for years. I doubt that this will ever be finished.
Maybe I will find another way to implement this. I will have a look.

[NickM-27]

I imagine mobilenetv2 would be good to support at a minimum for a built in model and parity with other detectors.

Unfortunately this is for the old generation of socs (not supported by Frigate and used by fewer people). Moreover, this has been in Preview state with no progress for years.

Dang that's unfortunate