zh/datasets/obb/

[giscus[bot]]

zh/datasets/obb/

了解用于Ultralytics YOLO 模型的 OBB 数据集格式。了解它们的结构、应用和格式转换，以增强物体检测训练。

https://docs.ultralytics.com/zh/datasets/obb/

[QLGalaxy]

Hi！I have transformed my data format from DOTA to yolo after spliting to 1024x1024. And I want to predict on the 1024*1024 test imgs, combine them and convert the format for submission to the official website for testing. Is there any tools for this?

[glenn-jocher]

@QLGalaxy to combine predictions and convert them for submission, you can use post-processing tools to merge the 1024x1024 tiles back into the original image format. While Ultralytics doesn't provide a specific tool for this, you might consider using image processing libraries like OpenCV or PIL to achieve the desired output. For more details on dataset handling, visit our documentation at https://docs.ultralytics.com/datasets/.

[QLGalaxy]

Thanks for your explaination. And further, I wonder how many GPUs do you use on training yolov11n-OBB.pt？

[glenn-jocher]

The number of GPUs used for training YOLO11n-OBB.pt can vary based on resources and requirements. Typically, multi-GPU setups are used to speed up training, but a single GPU can also suffice depending on the dataset size and model complexity. For specific configurations, you might want to experiment based on your hardware capabilities.

[UltralyticsAssistant]

👋 Hello, thank you for reaching out and showing interest in Ultralytics 🚀! Our Documentation is a great place to explore for detailed information on datasets and formats, including the OBB dataset format you're interested in. Be sure to check out the OBB section for insights on their structure, application, and format conversion.

If this is a 🐛 Bug Report related to datasets, please provide a minimum reproducible example to help us investigate further.

For any custom training ❓ Questions, including dataset details or model-specific queries, please provide additional context such as dataset samples and training logs. Also, consider reviewing our Training Tips for best practices.

Join the Ultralytics community for more interactive discussions. You can chat with us in real-time on Discord 🎧, explore in-depth topics on Discourse, or engage with others on our Subreddit.

Upgrade

Ensure you have the latest ultralytics package and all its requirements installed in a Python>=3.8 environment with PyTorch>=1.8:

pip install -U ultralytics

Environments

YOLO can be run in various environments with pre-installed dependencies:

• Notebooks with GPU:
• Google Cloud: See GCP Quickstart
• AWS: See AWS Quickstart
• Docker: See Docker Quickstart

Status

This badge indicates the status of our Continuous Integration (CI) tests, ensuring the correct operation of all YOLO Modes and Tasks across different platforms.

This is an automated response, but an Ultralytics engineer will be with you soon to provide further assistance. 😊

[LinkZY3471]

I encountered an issue while training on my custom dataset. In my dataset, some of the Oriented Bounding Boxes (OBB) are large and rotated, causing some corner coordinates to exceed the original image boundaries. This leads to values greater than 1 or less than 0 when converting to YOLO format, resulting in these samples being discarded during YOLO model training.
How should I handle these OBB boxes that exceed image boundaries? What's the recommended approach to process these cases while maintaining the integrity of the training data?

[glenn-jocher]

@LinkZY3471 to handle Oriented Bounding Boxes (OBBs) that exceed image boundaries, clip the corner coordinates to the valid [0, 1] range while ensuring the bounding box integrity. This adjustment prevents the loss of samples during training. For further details on managing OBB datasets, refer to the Ultralytics OBB guide.

[LinkZY3471]

Thank you, but if we simply clip the out-of-range values to 0-1, can the OBB shape still be correctly maintained? Would this cause any problems during training?

[glenn-jocher]

@LinkZY3471 clipping OBB coordinates to the [0, 1] range may slightly distort the box shape, especially if significant portions extend beyond the image boundary. However, this ensures all annotations remain valid for training. If preserving the exact OBB shape is critical, you could preprocess the data to crop or adjust annotations while maintaining their integrity. For details on handling OBB datasets, refer to Ultralytics OBB documentation.

[LinkZY3471]

Thank you :)

[glenn-jocher]

@LinkZY3471 you're welcome! If clipping affects OBB integrity significantly, consider preprocessing to adjust or split overly large boxes to fit within image bounds while maintaining their shape. For further insights, refer to the Ultralytics OBB guide. Let us know if you need additional clarification!