Add new task: Relationship Detection 🚀 #16411
Comments
|
👋 Hello @Maelic, thank you for your interest and for bringing this exciting idea to our attention 🚀! We're thrilled to hear about the potential new task of Relationship Detection using YOLOv8. This is an automated response, but rest assured an Ultralytics engineer will review your suggestion and provide feedback soon! For new feature proposals like this one, detailed information can greatly aid the discussion. If you have any additional insights or prototypes, please feel free to share them. Your linked work at SGG-Benchmark sounds promising! In the meantime, check out our Docs for any existing functionalities and our Tips for Best Training Results. Engage with the Ultralytics community through these channels:
UpgradeStay updated with the latest features by upgrading your pip install -U ultralyticsEnvironmentsTry running YOLOv8 in these verified environments:
Status
If this badge is green, all Ultralytics CI tests are passing, indicating that everything is operating smoothly. We look forward to exploring this potential feature further. Thank you for your initiative and willingness to contribute! 🙌 |
|
@Maelic thank you for your suggestion on Relationship Detection. While we appreciate your interest and initiative, we currently focus on maintaining and enhancing existing tasks. However, we encourage community contributions and welcome you to submit a PR if you wish to develop this feature further. |
Hi @glenn-jocher, thank you for your response. I would be happy to contribute to a PR, however implementing this new task in the codebase will require a lot of work: in addition to new model architectures, it requires changing the dataloader, and adding new evaluation methods for instance. For now, I won't have time to do it by myself. However, if someone in the community or in your team wants to help, I would be happy to work on a sub-part of the PR. |
|
Thank you for your willingness to contribute, @Maelic. While our team is currently focused on other priorities, we encourage community members to collaborate on this initiative. Feel free to reach out if you decide to work on a specific aspect of the PR. |
|
👋 Hello there! We wanted to give you a friendly reminder that this issue has not had any recent activity and may be closed soon, but don't worry - you can always reopen it if needed. If you still have any questions or concerns, please feel free to let us know how we can help. For additional resources and information, please see the links below:
Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed! Thank you for your contributions to YOLO 🚀 and Vision AI ⭐ |
Search before asking
Description
Dear @glenn-jocher and other members of the amazing Ultralytics team,
I would like to propose a new task for the YOLOv8 codebase: Relationship Detection.
Relationship Detection aims at predicting relations between objects in images or videos. This task is also called Scene Graph Generation (SGG) as the final output can be represented as a graph structure, grounded to the image with object bounding box coordinates.
Relationship Detection requires first predicting objects (bounding boxes and labels) and then predicting pairs and relations.
Relations are expressed in natural language and can be spatial relations (behind, in front of etc) or semantic relations (eating, jumping over etc).
Here is an example of the task:
Recently, SGG has gained a growing interest in real-time and low-resource approaches, which are lacking as of now.
To solve this gap, I have proposed an SGG codebase based on YOLOv8, V9 and V10, see https://github.com/Maelic/SGG-Benchmark.
I believe my work, or part of it, could serve as a baseline to build the Relationship Detection task directly in the Ultralytics codebase.
Use case
Use cases for real-time relationship detection are numerous:
Additional
I am just doing this issue to see if the community and the Ultralytics team would be interested in this feature.
If this is the case, I would be happy to contribute in implementing it.
Best
Are you willing to submit a PR?
The text was updated successfully, but these errors were encountered: