weird obb result and train batch about ultralytics HOT 8 OPEN

Hello @kwang19113,

Thank you for pointing out the similarity to the issue you've encountered before. It's helpful to cross-reference these cases to better understand potential underlying patterns or recurring problems.

Since setting all augmentation values to zero didn't resolve the issue, it might be worth exploring other aspects such as the data preprocessing steps or the model configuration. Sometimes, subtle nuances in how data is prepared or how the model is set up can lead to unexpected behaviors.

If you have any more insights or specific settings that you've tried since then, sharing those could be beneficial for further troubleshooting.

from ultralytics.

👋 Hello @kwang19113, 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.

from ultralytics.

Hello!

Thank you for providing detailed information and images regarding your issue with the OBB model for license plate detection. It seems like the misalignment in the bounding boxes during training might be affecting your final results.

Given the settings you've used, here are a couple of suggestions:

1. Augmentation Adjustments: The degrees=20 setting for rotation might be too aggressive, causing some bounding boxes to misalign. Try reducing this value to see if it improves the bounding box accuracy during training.

2. Batch Size: A batch size of 64 is quite large, especially if your GPU memory allows. Sometimes, reducing the batch size can lead to more stable training, particularly for complex tasks like OBB. Consider experimenting with smaller batch sizes if feasible.

3. Post-Processing: Ensure that your post-processing steps during prediction align with how your training labels are formatted and processed. Any discrepancies here could lead to results like those you're seeing.

If these adjustments don't resolve the issue, it might be helpful to look into the specific transformations applied to your training data and ensure they are correctly handled during the bounding box calculations.

Keep us updated on your progress, and feel free to reach out if you have more questions!

from ultralytics.

Hello!

Thank you for providing detailed information and images regarding your issue with the OBB model for license plate detection. It seems like the misalignment in the bounding boxes during training might be affecting your final results.

Given the settings you've used, here are a couple of suggestions:

1. Augmentation Adjustments: The degrees=20 setting for rotation might be too aggressive, causing some bounding boxes to misalign. Try reducing this value to see if it improves the bounding box accuracy during training.
2. Batch Size: A batch size of 64 is quite large, especially if your GPU memory allows. Sometimes, reducing the batch size can lead to more stable training, particularly for complex tasks like OBB. Consider experimenting with smaller batch sizes if feasible.
3. Post-Processing: Ensure that your post-processing steps during prediction align with how your training labels are formatted and processed. Any discrepancies here could lead to results like those you're seeing.

If these adjustments don't resolve the issue, it might be helpful to look into the specific transformations applied to your training data and ensure they are correctly handled during the bounding box calculations.

Keep us updated on your progress, and feel free to reach out if you have more questions!

Thank for the advice! i'll try to adjust accordingly and get back to you tomorrow with the result. Cheers

from ultralytics.

@kwang19113 could be related to this issue I reported a while back #10181

from ultralytics.

@kwang19113 could be related to this issue I reported a while back https://github.com/ultralytics/ultralytics/issues/10181

yea i think that's the one i set all augment value to 0 but the problem still persists

from ultralytics.

Hi @glenn-jocher

I'll try to look into it. and provided any helpful information for the team. I think I've encountered this problem with another task but Im not sure which one. I'll get back to you when i have any other relevant information.

Cheers

from ultralytics.

Hi @kwang19113,

Thank you for looking into this and for your willingness to share any relevant information. Your insights could be invaluable in resolving this issue. We appreciate your efforts and look forward to your update.

from ultralytics.