Yolov5 model making no detections

[Amna26103]

I am working on training a model using custom dataset in yolov5 with pytorch on vscode I was using ultralytics git repository
The model is executing without any errors train/exp were formed perfectly and the detections in validation batches were accurate

The program to load model is also executing without errors but it makes no detections

I tried using imgs,video and real Time the camera loads but doesn't makes any detections

Kindly guide me in this aspect

[sila-a0]

Hello, mine works in colab, Do you make it in colab?

[Amna26103]

Yup tried it but didn't work

[sila-a0]

!python detect.py --weights /content/yolov5/yolov5/runs/train/exp6/weights/best.pt --img 640 --conf 0.25 --source /content/yolov5/try.jpg I use this line of code for testing

[Amna26103]

It making no detections

[glenn-jocher]

@Amna26103 hello! It seems you're working with YOLOv5, but as you've reached out in the YOLOv8 repo, I'll guide you towards the right direction for YOLOv8.

For YOLOv8, if you're experiencing issues with no detections, it's important to ensure that your model is correctly trained and the weights are properly loaded. Also, check that your confidence threshold is not set too high, which might cause valid detections to be filtered out.

Since you've mentioned that the model works in Colab but not in your local environment, this could be an environment-related issue. Make sure that your local setup matches the one in Colab in terms of library versions and dependencies.

For further assistance, please refer to our documentation at https://docs.ultralytics.com, which provides comprehensive guidance on using YOLOv8, including troubleshooting tips that might be relevant to your situation.

If you continue to face issues, consider opening an issue in the YOLOv5 repository with detailed information about your training setup, validation results, and the exact command you're using for detection. The community there will be better suited to help you with YOLOv5-specific problems. Good luck! 🚀

[sila-a0]

Is the best.pt file created?

[Amna26103]

Yeah i tried using various dataset due to some reason my model was not working over this small dataset I changed the num to around 400 it started making detection but still not accurate thanks alot for helping though

[sila-a0]

I hope it gets even better

[glenn-jocher]

@ENNURSILA i'm glad to hear that your model has started making detections after increasing your dataset size. Achieving accurate detections often requires a combination of a sufficiently large and diverse dataset, proper model tuning, and adequate training time. If the accuracy is not yet up to your expectations, consider the following steps:

1. Data Quality: Ensure your dataset is well-labeled and representative of the use-case scenarios.
2. Hyperparameters: Experiment with different hyperparameters, which can significantly impact model performance.
3. Augmentations: Use data augmentations to increase the diversity of your training data without needing to collect more data.
4. Training Duration: Sometimes, models need more epochs to converge, especially with complex datasets.
5. Model Architecture: If you're still facing issues, you might want to try a different model architecture that could be more suited to your specific task.

Remember, model training is an iterative process, and improvements can often be made with careful analysis and adjustments. Keep experimenting, and best of luck with your YOLOv8 model! 🚀