Training YOLONAS from scratch for using it in a commercial application

[albertofernandezvillan]

As far as I understand for the license, if I used pre-trained weights for training YOLONAS (fine-tune my model on my dataset), I cannot use it in commercial applications. Is it right?

If so, when training:

from super_gradients.training import models
model = models.get('yolo_nas_l', 
                   num_classes=len(dataset_params['classes']), 
                   pretrained_weights="coco"
                   )

how to change this and train it from scratch?

[dagshub[bot]]

Join the discussion on DagsHub!

[NatanBagrov]

Hello, to get a model without pre-trained weights, you should remove the relevant part (pretrained_weights="coco"). This is applicable to any model.

For more information regarding the API, please read the docs, here is the relevant part

[albertofernandezvillan]

Thanks, and what about "I used pre-trained weights for training YOLONAS (fine-tune my model on my dataset), I cannot use it in commercial applications. Is it right?"

[shaydeci]

@albertofernandezvillan You are right.

[NeovMs]

If we use the pre-trained models for knowledge distillation to pre-train the Yolo-NAS from scratch, we're not strictly using the weights. In this case, can we get round the licence ?