Parameters Fusion

[znmzdx-zrh]

Search before asking

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

Question

How to integrate some parameters from imported external modules into the entire YOLOv5 model for joint training?I want to introduce some filters as a module into the YOLOv5 model to enhance images. Input the original image of Yolov5 to the result of additional enhancement module, and the enhanced image is obtained in the first layer of the convolution block into Yolov5, and then trained together.How can I merge the parameters inside the filters into the trainable parameter list of YOLOv5 for joint training and updating?Thank you for help.
In common.py

In yaml

In train.py

Additional

No response

[github-actions]

👋 Hello @znmzdx-zrh, 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):

• 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 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]

Hello,

Thank you for your question! To integrate external parameters and modules into the YOLOv5 model for joint training, you can follow these steps:

1. Modify the Model Architecture: Incorporate your enhancement module into the YOLOv5 architecture. This involves editing the common.py file to include your custom layers or modules.

2. Update the YAML Configuration: Ensure that your model configuration YAML file reflects the changes. Add the new layers or modules to the appropriate sections.

3. Adjust the Training Script: Modify train.py to include the parameters of your enhancement module in the optimizer. This ensures that the parameters are updated during training.

Here’s a brief example to illustrate:

In common.py:

class EnhancedModule(nn.Module):
    def __init__(self):
        super(EnhancedModule, self).__init__()
        # Define your enhancement layers here

    def forward(self, x):
        # Apply enhancement
        return enhanced_x

# Add your module to the YOLOv5 model
class Model(nn.Module):
    def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None, anchors=None):
        super(Model, self).__init__()
        self.enhanced_module = EnhancedModule()
        # Existing YOLOv5 layers

In your YAML configuration:

# Add your module configuration
enhanced_module:
  type: EnhancedModule
  args: []

In train.py:

# Include the parameters of the enhancement module in the optimizer
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.9, weight_decay=5e-4)

By following these steps, you can integrate your custom enhancement module into the YOLOv5 model and ensure its parameters are included in the training process.

For more detailed guidance, you can refer to the model ensembling tutorial.

Best of luck with your project! If you have any further questions, feel free to ask.

[znmzdx-zrh]

@glenn-jocher Your guidance is greatly appreciated!!I am studying your yolov5 project, and the problem of parameter fusion has puzzled me for a long time.Thank you very much for replying to my email.I have tried the method you mentioned》

[glenn-jocher]

Hello,

I'm glad to hear that the guidance provided was helpful to you! If you have any more questions or need further clarification as you continue working with the YOLOv5 project, please don't hesitate to reach out. We're here to help you make the most out of your experience with the model. Happy coding! 🚀

[znmzdx-zrh]

@glenn-jocher Hello,
I appreciate your help.But now I have a problem, the model in the verification phase output Labels, P, R, mAP are 0.
First of all, after your suggestion, I successfully added the parameters of the enhancement module into SGD training.As shown in the figure below, the 10 new weights and biases are the parameters of the enhancement module.



Then the training began.In first round of model training, it was found that the Labels and mAP output in the verification stage were 0. So I used the print function in the val.py file to see if the output of the validation phase model was correct and found that the output of the validation phase model is shown below.


The normal YOLOv5 model output should look like this.

I would like to ask you what is the reason for this situation? Whether the model did not correctly detect the target during the validation phase. I adjusted the IoU to 0.1, and again the output was 0. I am looking forward to your reply. Thank you very much.

[glenn-jocher]

Hello,

Thank you for providing detailed information about the issue you're encountering. It seems like the integration of the enhancement module might be affecting the model's ability to correctly process and detect objects during the validation phase.

Here are a few steps you can take to troubleshoot and potentially resolve this issue:

1. Check the Output of the Enhancement Module: Ensure that the output from your enhancement module is correctly formatted and scaled. An incorrect output format or scale might disrupt the subsequent layers of the YOLOv5 model.

2. Verify Integration Points: Revisit the points where the enhancement module integrates with the YOLOv5 architecture. Ensure that there are no dimension mismatches or data type inconsistencies.

3. Model Grad Check: Temporarily disable the enhancement module and run validation to check if the base YOLOv5 model without modifications is performing as expected. This can help isolate the issue to the enhancement module.

4. Hyperparameter Tuning: Since you've added new parameters, consider revisiting your training hyperparameters. Sometimes, the learning rate or other parameters might need adjustment to accommodate the new model complexity.

5. Debugging Outputs: Utilize debugging statements in your model's forward pass to log shapes and statistics of tensors passing through the network, especially at the junction points of your enhancement module and the main YOLOv5 architecture.

If these steps do not resolve the issue, it might be helpful to provide more details about the enhancement module's architecture and its output characteristics. This additional information could offer more insights into potential misconfigurations or errors.

[znmzdx-zrh]

@glenn-jocher Thank you very much for your reply. I will experiment the scheme you mentioned at once. I hope it will go smoothly.

[glenn-jocher]

@znmzdx-zrh hello,

You're welcome! I'm glad to hear that you're moving forward with the suggestions. I hope they prove helpful and that your experiments yield positive results. If you encounter any further issues or have more questions, feel free to reach out. Good luck! 🚀

[github-actions]

👋 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:

• Docs: https://docs.ultralytics.com
• HUB: https://hub.ultralytics.com
• Community: https://community.ultralytics.com

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 ⭐