search

ultralytics / yolov5

YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite
https://docs.ultralytics.com
GNU Affero General Public License v3.0
48.36k stars 15.87k forks source link

Confusion Matrix #13035

Open hsaine opened 1 month ago

hsaine commented 1 month ago

Search before asking

Question

Hello, I am working on a fence detection project with two classes: "fence" and "fence anomaly". Additionally, there is a third class, "background", which appears by default in my model. I want to enhance the performance of the model and I am curious about the expected percentage range for the "background" class. Is it normal for the "background" class to appear or not?

I already augmented the data with different operations and labeled the entire images using bounding boxes. Which parameters should I modify? I played with epoch time, batch size, and optimizer, but I didn't see any results.

confusion_matrix

Additional

No response

github-actions[bot] commented 1 month ago

👋 Hello @hsaine, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

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.

Requirements

Python>=3.8.0 with all requirements.txt installed including PyTorch>=1.8. To get started:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

Status

YOLOv5 CI

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Introducing YOLOv8 🚀

We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀!

Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects.

Check out our YOLOv8 Docs for details and get started with:

pip install ultralytics
glenn-jocher commented 1 month ago

@hsaine hello! 😊

It's great to hear about your project on fence detection. Regarding your question about the "background" class, it's quite common for object detection models like YOLOv5 to include a "background" class implicitly, especially to help the model distinguish between the foreground classes ("fence" and "fence anomaly") and other irrelevant parts of the image.

If you're looking to improve model performance and have already experimented with data augmentation, epochs, batch size, and optimizers without success, you might consider adjusting the learning rate or experimenting with different anchor box sizes which can be crucial for detecting objects of varying dimensions accurately.

Additionally, reviewing the loss components (classification, objectness, and box regression) during training might provide insights into what aspect of the learning process needs more tuning.

Keep up the good work, and don't hesitate to reach out if you have more questions! 🚀

hsaine commented 1 month ago

thank you for your response, Is it normal if it appears in high percentage in the matrix?

glenn-jocher commented 1 month ago

Hello @hsaine! 👋

Yes, it's fairly normal for the "background" class to appear in a high percentage in the confusion matrix, especially if a significant portion of your images consists of areas without the target objects ("fence" and "fence anomaly"). This indicates that your model is learning to identify non-target areas effectively, which is crucial for reducing false positives.

If the high percentage is a concern, consider checking if the bounding boxes in your training data are accurately and tightly defined around your target objects. Mislabeling or overly generous bounding boxes can lead to higher background predictions.

Keep up the good work! 😊

hsaine commented 1 month ago

Hello,

I am so sorry if I disrupt you. Now I am concentrating only on fence anomalies labeled as broken. If I want to decrease the false negatives, what should I do? Thank you! confusion_matrix

glenn-jocher commented 1 month ago

Hello @hsaine,

No disruption at all! 😊 To decrease false negatives for your "broken fence" class, consider the following strategies:

  1. Data Augmentation: Increase the variety and quantity of your training data with augmentations like rotations, flips, and brightness adjustments.
  2. Class Imbalance: Ensure your dataset has a balanced representation of the "broken fence" class.
  3. Hyperparameter Tuning: Experiment with different learning rates, batch sizes, and epochs.
  4. Anchor Boxes: Adjust anchor box sizes to better fit the dimensions of your "broken fence" instances.
  5. Model Architecture: Try different YOLOv5 models (e.g., YOLOv5s, YOLOv5m) to see which performs best.

Feel free to reach out if you have more questions. Good luck with your project! 🚀

github-actions[bot] commented 1 week 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 ⭐