WARNING ⚠️ NMS time limit 2.100s exceeded

[lqh964165950]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

为什么训练的每一轮都会出现这个警告？该怎么解决这个问题。

Additional

No response

[UltralyticsAssistant]

👋 Hello @lqh964165950, thank you for your interest in YOLOv5 🚀!

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug the issue, such as your system details, training parameters, and code snippet. Reproducing the issue on a smaller dataset can also be helpful.

If this is a custom training ❓ Question, please include as much information as possible, such as training arguments and logs, and ensure you are following best practices for training YOLOv5 models.

Requirements

Please ensure you have Python >= 3.8.0 and all required dependencies installed, including PyTorch >= 1.8. Verify your setup by cloning the YOLOv5 repository, navigating to its directory, and running the relevant scripts.

Environments

YOLOv5 can be run in diverse environments, including Jupyter notebooks, cloud services like Google Cloud or AWS, or even Docker containers. Make sure your runtime is properly configured and equipped with the necessary dependencies like CUDA, cuDNN, Python, and PyTorch.

Status

Our Continuous Integration (CI) system verifies the correct operation of YOLOv5 across various processes, such as training, validation, inference, export, and benchmarks, and helps ensure its reliability on Windows, Ubuntu, and MacOS.

This is an automated response to help you get started. An Ultralytics engineer will follow up with you soon to assist further. Let us know if you can provide additional details to help us understand and address your issue. 😊🚀

[pderrenger]

@lqh964165950 thank you for reaching out. The "NMS time limit exceeded" warning indicates that the Non-Maximum Suppression (NMS) step during inference is taking longer than expected, potentially due to a high number of predictions per image. To address this:

1. Try reducing the number of predictions by increasing the conf-thres value (e.g., --conf 0.5) in your training or inference command.
2. Ensure you’re using the latest version of YOLOv5 and the Ultralytics package as updates may resolve performance issues. Update with pip install --upgrade ultralytics.
3. For further optimization, consider lowering the image size (--img-size) or debugging for unusually complex images.

If the issue persists, feel free to share additional details on your training settings.