feat: add yolov10 node

[storrrrrrrrm]

Description

add yolov10 node

Related links

#8753

Parent Issue:

• Link

How was this PR tested?

Notes for reviewers

None.

Interface changes

None.

Effects on system behavior

None.

[github-actions]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[xmfcx]

Thank you very much for the contribution!

Please pass the tests marked as Required then when it is finished, you can mark it as Ready for review.

[storrrrrrrrm]

2024-11-04.16.59.28.mp4

[Shin-kyoto]

@storrrrrrrrm
[must]
Can you use Apache 2.0?
Did you check the licence of Yolov10? It is AGPL3.0

cf. https://www.reddit.com/r/computervision/comments/1e3uxro/ultralytics_new_agpl30_license_exploiting/?rdt=41888

both the training code and the models produced by that code are covered by AGPL-3.0.

[liuXinGangChina]

Morning Shin san @Shin-kyoto ， the cf link you provide can not be open due to net block policy

I read the AGPL3.0 license，which is pretty strict when code used for commercial purpose，at the same time it may also contaminate the project。While Apache 2.0 looks more friendly。

for the inference code ，since we reference the original code from tire4‘s tensorrt-yolo-x， so i believe it should follow the tire4 license。 And for the model and retrain code，we are not sure how to deal with。

I have an idea and would like to discuss with you

1. create a tensorrt-yolo-v10 node in autoware universe repo using apache 2.0。 Highlight a notice in the readme file which tell the developer that model and retrained model under AGPL 3.0， those who want to use the model in their project shall consider the license issue seriously
2. and if 1 works ，we may fork the retrain code in a separate repo outside autoware and retrain the original model with tire4 dataset then release the fine-tune model for autoware developer use

What do you think about the above solution，feel free to hear comment from you

Have a nice day！
Xingang

[xmfcx]

Hi @Shin-kyoto and @liuXinGangChina 👋

Yes, this is just the inference code. And the inference code is based on autoware_tensorrt_yolox package which uses Apache 2.0. So this package can also make use of this license.

We should separate the training, model and inference licenses.

For now, @storrrrrrrrm could you update the readme file to include the information about the training code and the model licenses?

Example: https://github.com/autowarefoundation/autoware.universe/tree/main/perception/autoware_lidar_centerpoint#legal-notice