rockchip: update dependencies and add script for model conversion

[MarcA711]

Proposed change

This PR updates the following rockchip dependencies:

• FFmpeg 6.1.5 to 6.1.6 (includes a couple of bug fixes during transcoding)
• rknn-toolkit-lite2 from v2.0.0 to v2.3.0
• yolonas models were updated and converted using toolkit v2.3.0

Moreover, the image supports converting models from onnx to rknn format now. This should be useful for Frigate+ users who want to use their own models. Changes:

• the script conv2rknn.py was added
• python package rknn-toolkit2 was installed
• 20 images from the COCO dataset are bundled with the image (rknn-toolkit2 needs these during quantization).

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

Some notes about yolonas:
It seems the super-gradients project is not active anymore after Deci AI was bought by Nvidia. The current pypi release of super-gradients is broken because the download URL changed (see Deci-AI/super-gradients#2057). There are a couple of fixes:

1. Download the weights manually from the new url (https://sg-hub-nv.s3.amazonaws.com/models/yolo_nas_s_coco.pth) and save them under ~/.cache/torch/hub/checkpoints/./.cache/torch/hub/checkpoints/
2. Manually fix the package by following these instructions
3. Install super-gradients from this fork that merged the fixes.

This is relevant for users who try to use this notebook from the frigate repo:

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)

[netlify]

✅ Deploy Preview for frigate-docs ready!

Name	Link
🔨 Latest commit	c4e1191
🔍 Latest deploy log	https://app.netlify.com/sites/frigate-docs/deploys/676ed1b82be41c00089d9217
😎 Deploy Preview	https://deploy-preview-15699--frigate-docs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[NickM-27]

I think it would be best to target 0.16 with these changes, let me know what you think

[MarcA711]

I agree

[NickM-27]

Are there any limitations on what model architectures can be converted to rknn format?

[MarcA711]

On AArch64 only onnx to rknn is supported. This might change in the future.

On x86 other formats are available as well.

[NickM-27]

Sorry I mean are there any restrictions for example only yolonas, or also mobilenet and others?

[MarcA711]

Sorry, I misunderstood.

In theory one can convert any model. Practically it's not a straightforward process. Not all onnx operations are supported.

For example, the default yolo-nas model includes post-processing and performes a batched non maximum suppression. The rknn-toolkit2 doesn't support this, so I had to modify the model and do the NMS programmatically in python.