Continue training

[pjh11214]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

How to continue training on the basis of already trained models

Additional

No response

[UltralyticsAssistant]

👋 Hello @pjh11214, thank you for reaching out and showing interest in YOLOv5 🚀! Continuing training from a pretrained model is a common task and I'm here to help guide you. An Ultralytics engineer will also assist you soon.

To continue training with an already trained model, you generally need to specify the path to your weight file using the --weights argument when executing the training script. This typically looks something like this:

python train.py --weights path/to/your_weights.pt --cfg path/to/your_custom_config.yaml --data path/to/your_data.yaml --epochs 50

To ensure you've set everything up properly, here's a brief reminder about the environment requirements:

Requirements

Python>=3.8.0 with all requirements.txt installed including PyTorch>=1.8. You can set up your environment as follows:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

If this is a 🐛 bug report, please provide a minimum reproducible example to help us identify the issue. For questions involving custom training, supplying additional information like dataset examples and training logs can be immensely helpful.

Feel free to check out these helpful resources for more detailed guidance:

• Custom Data Training Guide
• Tips for Best Training Results

Explore our new YOLOv8 for advanced features and capabilities:

Introducing YOLOv8 🚀

We're thrilled to present our state-of-the-art model, YOLOv8! Designed for enhanced performance in object detection, segmentation, and classification tasks.

Get started with YOLOv8 by running:

pip install ultralytics

Remember, our community and resources are here to help you succeed with your machine learning endeavors! 🚀

[pderrenger]

@pjh11214 to continue training an already trained YOLOv5 model, use the --weights argument to specify your model's checkpoint file, like this: python train.py --data your_dataset.yaml --weights path/to/your_model.pt.