Angle change and atan2 function about ultralytics

[github-actions]

👋 Hello @ekthaliz, thank you for your interest in Ultralytics YOLOv8 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Join the vibrant Ultralytics Discord 🎧 community for real-time conversations and collaborations. This platform offers a perfect space to inquire, showcase your work, and connect with fellow Ultralytics users.

Install

Pip install the ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8.

pip install ultralytics

Environments

YOLOv8 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLOv8 Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@ekthaliz hello,

Thank you for your detailed questions regarding the angle adjustments in YOLOv8 OBB. Let's address each of your queries:

1. Angle Correction to 360°:
   To modify the angle range from 0° to 360°, you will need to adjust the angle calculation in the model's post-processing steps. The line angle = (angle.sigmoid() - 0.25) * math.pi can be adapted to cover the full 360° range. After making this change, ensure that any post-processing steps that rely on angle values are updated accordingly. This includes loss calculations and any angle-based filtering or adjustments.

2. Angle Range 45° to 135°:
   The angle range of 45° to 135° was chosen to simplify certain geometric calculations and ensure consistency in the model's predictions. The commented-out code angle = angle.sigmoid() * math.pi / 2 suggests an alternative approach to angle normalization, but it was likely commented out due to specific issues or inconsistencies observed during development. Using a restricted angle range can help avoid ambiguities in object orientation.

3. atan2 Function Usage:
   The atan2 function is used to compute the angle of a vector in a way that considers the full 360° range, providing a more comprehensive angle representation. This is particularly useful for practical applications where the full orientation of an object is needed. The reason for not setting the code to 180° initially could be to maintain simplicity and avoid potential issues with angle wrapping and discontinuities.

   • atan2 Function Location:
     The atan2 function might not be explicitly visible in the code if it is used indirectly through other libraries or functions. You can search for its usage in the codebase to understand where and how it is applied.

   • Modifying atan Function for 360°:
     If you are extending the angle range to 360°, you may need to replace any atan function calls with atan2 to ensure correct angle calculations. The atan function typically supports angles up to 90°, so using atan2 will provide the necessary support for the full 360° range.

To proceed, I recommend:

• Reviewing the code to identify all instances where angles are calculated or used.
• Updating the angle normalization and post-processing steps to handle the 0° to 360° range.
• Ensuring that any loss functions or evaluation metrics that depend on angle values are adjusted accordingly.

If you encounter specific issues or need further assistance, please provide a minimum reproducible example of your code changes. This will help us better understand the context and provide more targeted support. You can find guidelines for creating a minimum reproducible example here.

Feel free to reach out with any additional questions. Happy coding! 😊