img-weights doesn't seem to work

[liquored]

Search before asking

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

Question

I want to train a model on my own data, but there is a certain gap in the data. Because of that, I use the trick image-weights, but it doesn't seem to have much effect. The fewer amounts of data get lower precision, recall and map.
Could you please give me some advice? I have tried OHEM, which makes my model divergent.
Look forward to your reply!

Additional

No response

[github-actions]

👋 Hello @liquored, 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>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. 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

[liquored]

I am sorry, maybe I didn't make sense. At first, I make the parameter 'ima-weights' False, and I found the categories which have fewer numbers get worse precision, recall and map. Then I make the parameter 'ima-weights' True, but it doesn't make any change of precision, recall and map.

[liquored]

I have already solved this problem. I am sorry, I made a mistake last week. After my analysis and experimentation, the parameter 'ima-weights' works well. My problem is because the target is too small compared to other categories. While I make the input size bigger, the precision, recall and map have improved a lot.

[glenn-jocher]

@liquored, thank you for updating the community and sharing your solution. I'm glad to hear that you were able to resolve your issue with the 'ima-weights' parameter and improve your model's performance by adjusting the input size. It's great to see that you're experimenting with different parameters and techniques to improve your results, and I wish you continued success with your YOLOv5 projects. If you have any further questions or issues in the future, please don't hesitate to ask!

[liquored]

@liquored, thank you for updating the community and sharing your solution. I'm glad to hear that you were able to resolve your issue with the 'ima-weights' parameter and improve your model's performance by adjusting the input size. It's great to see that you're experimenting with different parameters and techniques to improve your results, and I wish you continued success with your YOLOv5 projects. If you have any further questions or issues in the future, please don't hesitate to ask!

@glenn-jocher Yes, I have resolved my issue, but in fact, there is still a question why the OHEM does't work. Not only that, but it also seems to make the model worse, just like the focal loss which also makes my model worse.

[glenn-jocher]

@liquored, it's great to hear that you found a solution to your initial problem. As for OHEM and focal loss, they can be powerful tools when applied correctly, but they may not always yield positive results in every scenario. If you need further assistance with these techniques or encounter any other challenges, feel free to ask for help. Your perseverance and dedication to improving your YOLOv5 model are commendable, and I'm here to support you in any way I can. Keep up the great work!