Yolov5: How to decrease false positives ?

[Zegorax]

❔Question

Recently YOLO started to give me a lot of incorrect results and I thought that removing the annotation and adding the image (without .txt, blank txt or with other correct annotations) to the training set would make it learn there is nothing there, but it is not working.

What is the best way to decrease false positive rate ? How would you do it ?

[github-actions[bot]]

Hello @Zegorax, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[glenn-jocher]

@Zegorax you control your own FP rate by setting --conf to whatever you want at inference time. Precision is completely up to you, it's not a property of the trained model.

[Zegorax]

@glenn-jocher Hello Glenn, thank you for answering me.

I'm trying to improve the detection, because those false positive have sometimes a high confidence level.

I'm wondering what I can do to improve the model and make it not detect / predict those false positives

[glenn-jocher]

@Zegorax then you should use a larger model, more training data, increase resolution, try different augmentation, vary hyperparameters, etc.

Nothing new here, just the usual suspects for improving results.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Zegorax]

@glenn-jocher Thank you. I improved the training set a little bit with more data. What I was wondering is the following : Is there a way to give YOLO an image without annotation and make him learn that there's nothing on this image ?

[Ownmarc]

@Zegorax yes, no annotation file is needed when there is nothing in the image. The image will still be used to train on as any other image in the train directory. I personally prefer to create an empty annotation file just so my image count always match my annotation file count.

[Zegorax]

@Ownmarc That's perfect then, it's what I've been doing the whole time. Thank you very much!

[wwdok]

@Zegorax yes, no annotation file is needed when there is nothing in the image. The image will still be used to train on as any other image in the train directory. I personally prefer to create an empty annotation file just so my image count always match my annotation file count.

Hi, What do you think the ratio of the the dataset with objects and dataset without objects should be ？