CVHub520
Ultralytics Training Enhancements: Dataset Split, YAML & Model Export
Search before asking
- [x] I have searched the X-AnyLabeling issues and found no similar feature requests.
Description
1.ultralytics 训练的时候 数据集划分不是很理解，前面有yaml 数据集，然后多一个划分比例。这在分割数据集上产生了混乱。其次如果是训练打开的数据集，还需要yaml文件干啥，就显得十分的混乱 2.数据格式标签像yolo这种,可以连同yaml一起导出，免得麻烦还得自己写 3.ultralytics 训练功能 到处权重可以，到根目录供后续打标签，也可以用来推理，所以这个应该在任务开始有个选择，是用来智能标注还是为了后续使用 感谢作者大大的贡献
Use case
1
Additional
1
Are you willing to submit a PR?
- [x] Yes I'd like to help by submitting a PR!
Hello qinxue123321, 👋
Thanks for your valuable feedback and detailed suggestions! We truly appreciate you taking the time to share your thoughts on the Ultralytics training feature. Let me address your points one by one:
-
Regarding the dataset split and YAML file for Ultralytics training: The Ultralytics training component in X-AnyLabeling currently defaults to loading all data from the active task. It then performs the train/validation split based on the ratios you've set within the interface. The YAML file is required to maintain consistency with the official Ultralytics training process, which aims to minimize the learning curve for users already familiar with their ecosystem.
-
On the data format and YAML export: Please note that Ultralytics is one of many integrated features within X-AnyLabeling, and the platform isn't solely designed around it. While you might need to create a YAML file once, it's typically reusable across subsequent tasks. For initial setup, we recommend downloading an official Ultralytics YAML template and modifying it to fit your specific needs.
-
For exporting weights and intelligent labeling: For intelligent labeling, you can absolutely follow the official steps for adding custom models by simply setting the appropriate
onnxmodel directory path. There's no need to modify or move exported files for this purpose. The weights can indeed be used for subsequent labeling or inference directly from the task's output directory.
Finally, it's important to clarify that X-AnyLabeling is designed as an open, all-in-one platform for training, inference, and labeling, rather than being specifically tailored for Ultralytics alone. Our design process aims to consider and accommodate a wide range of use cases and scenarios, rather than optimizing for one specific user's requirements.
As an open-source project, we strongly encourage you to customize it deeply based on your actual usage to unlock its maximum potential. We really appreciate your willingness to submit a PR – that's fantastic! 🙌
Best regards, X-AnyLabeling Maintainer
