Male Female Detection

[AJITKUMAR130012]

I have finetuned the yolov11 for the male and female dataset. For each male is is giving the good result, but for the female it is predicting male some times. So how to make stable prediction:
Train size: 1431 images
total male: 5315
Total female: 2365

Because i have trained on the bus shelter, So there are less female present in the images.

So it it imbalance dataset, Thats why i am getting the bad result for women.

[UltralyticsAssistant]

👋 Hello @AJITKUMAR130012, thank you for reaching out with your inquiry about male and female detection using YOLOv11 🚀!

It sounds like you're facing challenges with class imbalance in your dataset, which can indeed affect the model's performance. We recommend exploring strategies to address this, such as data augmentation or weighted loss, which you might find helpful. For more in-depth guidance, please refer to our Tips for Best Training Results.

If this is a 🐛 Bug Report, and the issue persists despite addressing the imbalance, we kindly ask you to provide a minimum reproducible example for further debugging support.

Additionally, ensure your environment is up-to-date by upgrading to the latest ultralytics package with:

pip install -U ultralytics

You can also run your experiments in one of our verified environments for optimal performance:

• Notebooks with free GPU:

For real-time conversations and support, join our Discord community 🎧, or if you prefer more detailed discussions, you can visit our Discourse forum. We're also on Reddit for broader community engagement.

This is an automated response to get you started, and an Ultralytics engineer will further assist you soon.

[glenn-jocher]

@AJITKUMAR130012 to improve predictions on an imbalanced dataset, consider augmenting your dataset with more female images or using techniques like class weighting during training. You can also explore data augmentation strategies to enhance model robustness. For further guidance, visit our YOLO Common Issues page.

[AJITKUMAR130012]

In my traing dataset, images are like that so total number of male is greater than the female. So if i will augment then Number of male increase propotionally( I think it will be more imbalance).?

So i will try with "techniques like class weighting during training"

[glenn-jocher]

To address the imbalance, consider techniques like class weighting or oversampling the minority class. Augmentation can help, but ensure it balances both classes. For more tips, visit our YOLO Common Issues Guide.