Pre-training weights for YOLOv8 multispectral target detection algorithm

[luckyboby12]

I have developed a multispectral target detection algorithm based on yolov8, how should I use the pre-training weights during model training?

[UltralyticsAssistant]

👋 Hello @luckyboby12, thank you for your interest in Ultralytics 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, such as yours, please provide as much information as possible to help us assist you better. For example, details about your dataset (including image examples), the pretraining weights you are using (if applicable), any modifications made to the YOLOv8 architecture, and your training logs would be very useful. Additionally, you can verify you are following our Tips for Best Training Results to optimize your process.

Join the Ultralytics community where it suits you best. For real-time chat about your project, join us on Discord 🎧. For thoughtful, ongoing discussions, visit Discourse. Or share your insights on our Subreddit!

Upgrade

Make sure your ultralytics package version is up to date to ensure compatibility and access to the latest features and improvements. You can upgrade using the following command (requires Python>=3.8 and PyTorch>=1.8):

pip install -U ultralytics

Environments

YOLOv8 can be run on various environments to suit your needs, whether you're training locally or using a cloud service:

• Notebooks with free GPU access:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

This is an automated response to help you get started 😊, but don't worry – an Ultralytics engineer will follow up soon to assist you further if needed!

[glenn-jocher]

@luckyboby12 to use pre-trained weights for your YOLOv8-based multispectral target detection algorithm, specify the weights file (e.g., yolov8n.pt) when loading the model in your training script or CLI command. For example, in Python: model = YOLO("yolov8n.pt"). You can then fine-tune the model on your custom dataset. For more details, refer to the Train Mode documentation.

[luckyboby12]

My network structure has changed a lot. Can I still use the pre-training weight of yolov8n.pt at this time

[glenn-jocher]

@luckyboby12 yes, you can use yolov8n.pt as pre-trained weights, but ensure your modified network's layers and dimensions are compatible with the weights to avoid mismatches. You may need to fine-tune or partially load weights for best results.

[luckyboby12]

I designed the feature extraction backbone network of yolov8 into a two-stream structure to meet the simultaneous loading of visible and infrared images.

[luckyboby12]

Hello ! In fact, I am not sure whether the effect of pre-training weight is effective, and sometimes the effect of using pre-training weight is not necessarily good. I want to know how the pre-training weight works.

[glenn-jocher]

Pre-trained weights provide a starting point for training by leveraging learned features from a large dataset, which can accelerate convergence and improve performance, especially with limited data. However, their effectiveness depends on the similarity between the pre-training dataset and your target dataset. Fine-tuning can help adapt these weights to your specific task.