Bounding box duplicated?

[LOCKminiumRSY]

so i tried to test my data by taking the picture of the object, and when the YOLO predict, it shows two bounding box, in one image, the label is the same while on the other image , the label is difference

the program supposed to predict if the table is occupied or not and it based on if the person is sitting on the chair behind or in front of the table and show only one of the bounding box with the predicted label.

the label is 'meja 2 kosong', 'meja 2 terisi", meja 3 kosong"
it translate to 'table 2 not occupied, table 2 occupied', table 3 not occupied'

i was searching for the explanation why it happened, but couldn't find it. i'll be great if anyone can explain to me why it

this is the picture of the predicted image

[UltralyticsAssistant]

👋 Hello @LOCKminiumRSY, thank you for sharing your observations and using Ultralytics 🚀! Your use case sounds intriguing, and visual predictions like this can indeed raise some unexpected behavior.

We recommend visiting the Docs for detailed insights and tutorials. If you're new to training or prediction outputs, check our Model Training Tips and CLI or Python inference guidance to ensure your usage aligns with best practices.

If this is a 🐛 Bug Report, please provide a minimum reproducible example (MRE). Examples would include your code snippet, model configuration, and relevant dataset details so we can better understand and debug the situation for you.

If this is more of a ❓ Question, kindly share additional details about your dataset (e.g., sample images if possible), your model (e.g., yolov8n.pt or a custom-trained model), and training logs. Your insights and setup context will help us and the community provide more precise feedback.

Upgrade

First, verify you are using the latest version of the ultralytics package. Upgrade in a Python>=3.8 environment with PyTorch>=1.8:

pip install -U ultralytics

This ensures your issue is not already resolved in the latest release.

Environments

Make sure your project is running in a supported environment. We provide verified setups where dependencies like CUDA, Python, and PyTorch are preinstalled:

• Notebooks with free GPU access:
• Google Cloud: GCP Quickstart Guide
• Amazon AWS: AWS Quickstart Guide
• Docker: Docker Quickstart Guide

Explore and Collaborate

For real-time discussions, you can join our awesome community on Discord 🎧. Prefer more in-depth conversations? Visit our Discourse forum. Or get involved in shared insights on Reddit.

Status

Finally, check the current Ultralytics CI for test results. If this badge is green, all tests on YOLO modes and tasks are passing across different platforms!

This is an automated response, but don't worry, an Ultralytics engineer will review your issue and assist you soon 😊

[glenn-jocher]

@LOCKminiumRSY duplicate bounding boxes can occur due to overlapping predictions that were not suppressed during Non-Maximum Suppression (NMS). Ensure your model's iou threshold for NMS is appropriately set to remove redundant detections. If the issue persists, verify your dataset annotations for consistency and alignment with your intended task. For more on bounding box postprocessing, refer to this section.

[LOCKminiumRSY]

alright i'll try to set iou threshold, thank you

[LOCKminiumRSY]

what's the default value for conf and IoU argument? when i check the website, the val is lower than predict

[glenn-jocher]

@LOCKminiumRSY the default conf value is 0.25 and the default iou value is 0.45 for predictions. These values can be adjusted as needed. For validation, the thresholds are stricter, often with an iou of 0.7 for more precise evaluation. You can reference the documentation for more details: Inference Arguments.

[LOCKminiumRSY]

got it thanks a lot

[glenn-jocher]

You're welcome! If you have any further questions, feel free to ask.