Getting no-detections error from YOLOv8

[SaiCraze]

Hello! So, I have successfully trained a model using YOLOv8 with correct annotations and images. I have put the correct data.yaml file as well. However, when I go ahead and upload ANY image or even video or webcam feedback, it gives me this: "no-detection". Please how to fix this?

btw, when I try with the default model, it works. Its just the custom model that doesn't work :(

[UltralyticsAssistant]

👋 Hello @SaiCraze, thank you for reaching out and for using YOLOv8 🚀! We appreciate you bringing this to our attention.

If this is a 🐛 Bug Report, it would be super helpful if you could provide a minimum reproducible example (MRE) along with more details such as:

• The exact command or code snippet you are using to run the model on your images/videos
• The custom model weights you trained (please confirm if these are saved in .pt format)
• Screenshots or logs generated during the training process—this helps identify if there were any training irregularities

If this is a custom training question ❓, please ensure:

1. Your images and annotations were correctly labeled and formatted for training. YOLOv8 uses the YOLO annotation format—verify this with Roboflow or another compatible tool.
2. You followed the Tips for Best Training Results, such as ensuring diverse datasets that cover edge cases for your object classes.
3. You checked the confidence threshold (conf) parameter during inference. Low thresholds might excessively filter detections.

For further debugging:

• Consider testing your model with the --save-txt or --save-conf flags during inference to output detections explicitly, even if non-visible in visualizations.
• Run inference in debug mode with verbose=True to obtain additional logs.

Join the Ultralytics community if you'd like real-time support or to brainstorm with other users:

• Chat with us on Discord 🎧
• Engage in deeper conversation on Discourse
• Participate in discussions in the Subreddit

Upgrade

Ensure your environment is up-to-date by upgrading to the latest ultralytics package and all requirements in a Python>=3.8 environment with PyTorch>=1.8:

pip install -U ultralytics

Environments

For optimal results, we recommend running YOLOv8 in one of the verified, up-to-date environments with all dependencies, such as:

• Interactive GPU-based notebooks:
• Cloud-based solutions: GCP Quickstart Guide or AWS Quickstart Guide
• Local installations: Use the Docker Image

Status

When this badge is green, all Ultralytics CI tests are currently passing, ensuring the framework's expected operation across all core functionality.

🔎 An Ultralytics engineer will review your issue soon and provide more targeted guidance. Stay tuned—thank you for your patience and for being part of our community! 😊

[Y-T-G]

It depends on how good your model is. If you trained it on just a few images, then it wouldn't be good.