Need help in Training a Cutom Model

[LasalJayawardena]

I want to build a YOLO model by training on a set of images (training set). Then save the best model by validation loss (Like Tensorflow Checkpointer). Finally, I want to use it for predictions.

How can I implement this? I am a bit unclear about the training phase and using the saved trained model. Any help is appreciated! Thank you in Advance.

[leeping-ng]

Hi! PeekingDuck is designed for model inference rather than model training, so I suggest finding a YOLO repo that comes with a training script to train your model separately. Subsequently, you can create a custom PeekingDuck node for your trained model by referring to this guide, and use PeekingDuck for inference. Our model.yolo_license_plate node was created using this method.

[leeping-ng]

Hi @LasalJayawardena, we have released a new tutorial on how you can use your own trained models with PeekingDuck. Hope that this helps with your project.