search

ultralytics / ultralytics

Ultralytics YOLO11 🚀
https://docs.ultralytics.com
GNU Affero General Public License v3.0
32.47k stars 6.24k forks source link

Yolo new features #6386

Closed Alarmod closed 10 months ago

Alarmod commented 12 months ago

Search before asking

Description

Can ideas from this paper be emplemented in Yolo? https://www.mdpi.com/1424-8220/23/16/7190

For small objects detection

Use case

No response

Additional

No response

Are you willing to submit a PR?

glenn-jocher commented 12 months ago

@Alarmod hello and thank you for your interest in contributing to the YOLOv8 development!

It's fantastic to see community members eager to expand YOLOv8's capabilities with new research and ideas. If you've identified methods from a recent paper that you believe could enhance the model's ability to detect small objects, we're certainly open to considering these for integration.

Please be aware that incorporating new features derived from academic papers involves a careful process. It includes understanding the proposed methods, evaluating their compatibility with the existing framework, and rigorous experimentation to ensure they improve the model without introducing negative side effects like decreased accuracy on other tasks or inflated computational costs.

Given that you've expressed a willingness to submit a PR, that's a great start, and much appreciated! If you move forward with this, make sure that the implementation is well-tested and provides clear improvements in the detection of small objects.

I recommend first opening a discussion on our GitHub repository where you can outline the specific improvements suggested by the paper. This will allow us to provide feedback and ensure that there's alignment with YOLOv8's roadmap before you invest time and effort into the coding part. Please include your experimental results, comparative analysis, and possibly ablation studies that validate the effectiveness of the proposed changes.

We look forward to seeing your innovation and how it can push the boundaries of object detection further. Your willingness to contribute is invaluable to the YOLO community and the continuous evolution of Ultralytics models.

Also, remember that our documentation at https://docs.ultralytics.com could provide additional guidelines on contributing to the project. It also outlines best practices for submission which you may find useful during the PR process.

Thank you once again for your initiative, and happy coding! 🚀

Alarmod commented 12 months ago

Maybe my graduate student will deal with these issues. I am his leader and interested also in nonregid objects true detection (smoke, fire, medical objects and so on). This objects don't have stable form and may to crumble into pieces.

glenn-jocher commented 12 months ago

@Alarmod, it's wonderful to hear that you and your graduate student are interested in the detection of non-rigid objects like smoke, fire, and medical elements, which certainly represent unique challenges in the field of computer vision due to their amorphous and dynamic nature.

YOLOv8 could potentially be adapted to better handle such variable shapes through customized training on datasets that are rich in these types of objects. As you're likely aware, the performance of deep learning models like YOLOv8 heavily relies on the quality and representativeness of the training data. For non-rigid object detection, ensuring your dataset includes a wide variety of these objects in differing states and overlaps could be crucial to enhancing detection accuracy.

For your graduate student's project, you could explore enhancing YOLOv8's model architecture or loss functions to make them more sensitive to the peculiar characteristics of non-rigid objects. You may also want to look into data augmentation techniques that can simulate the variability in shape and appearance that non-rigid objects experience.

As you embark on this research journey, don't hesitate to engage with the YOLO community. Discussion on how to approach these challenges and sharing preliminary findings can foster collaborative solutions and might prompt valuable feedback. Furthermore, if successful methods are developed, they could not only benefit your projects but also the wider research community and numerous real-world applications.

I recommend reading our documentation, which includes information on training models on custom datasets and other advanced topics that might support your endeavours in pushing the boundaries of what's possible with YOLOv8 when dealing with the complexities of non-rigid object detection.

Best wishes to you and your student on this promising venture, and we look forward to any contributions or insights you may be able to share with the Ultralytics YOLO community. Good luck, and may your research yield impactful results! 🌟🔥🧠

Alarmod commented 12 months ago

Another paper for Yolo new features: https://www.sciencedirect.com/science/article/pii/S0950705122000612

glenn-jocher commented 12 months ago

@Alarmod thank you for pointing us to additional research that might inform the ongoing development of YOLOv8. Research papers often contain valuable insights that can be harnessed to enhance performance or introduce novel capabilities to existing systems like YOLOv8.

As mentioned before, integrating ideas from academic papers into YOLOv8 requires:

  1. Critical evaluation to determine their relevance and potential impact on existing model performance.
  2. Assessment of the compatibility of new techniques with the YOLOv8 architecture and its objectives.
  3. Implementation of proof-of-concept modifications in a controlled environment.
  4. Rigorous testing and validation to quantify improvements and identify any drawbacks.

If after reviewing the paper, you or your graduate student believe there is a compelling case for incorporating its ideas into YOLOv8, I recommend the following steps:

If there is consensus that the proposed features could yield significant benefits, and you or your student is ready to work on a PR, we would be thrilled to see the results of your efforts. Remember that a successful PR should include not only the code but also comprehensive testing that demonstrates the feature's advantages and ensures it does not adversely affect other aspects of the model's performance.

Don't forget to make good use of the resources available in our documentation, which can be a helpful guide through the process of customizing and contributing to YOLOv8.

Engaging with cutting-edge research is a cornerstone of advancing machine learning models. We are excited about the potential to include your contributions in YOLOv8 and wish you success in your endeavors to improve the state-of-the-art in object detection. 🚀🔍📈

github-actions[bot] commented 11 months ago

👋 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 ⭐