search

WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 572 forks source link

İs there any model to apply transfer learning? #269

Open omeryasar opened 3 years ago

omeryasar commented 3 years ago

Hi, I want to train this repository for my custom objects. But as far I observed, I have to train in scratch in such a case. But rather than that I prefer to fine tune a pretrained model if its exist. So is there anyone following that approach? if so can you tell me how can i do it. Or is there any pretrained model exist for fine tuning.

Thanks in advance.

SashaAlderson commented 3 years ago

These models are trained on coco dataset to detect 80 classes. You can download weights of this models and finetune them to detect your custom objects. There are roboflow guide on youtube for scaledYOLOv4.

omeryasar commented 3 years ago

These models are trained on coco dataset to detect 80 classes. You can download weights of this models and finetune them to detect your custom objects. There are roboflow guide on youtube for scaledYOLOv4.

Hi first thanks for reply. What do you mean by these models?

SashaAlderson commented 3 years ago

These models are trained on coco dataset to detect 80 classes. You can download weights of this models and finetune them to detect your custom objects. There are roboflow guide on youtube for scaledYOLOv4.

Hi first thanks for reply. What do you mean by these models?

Download weights from https://github.com/WongKinYiu/ScaledYOLOv4 and you'll get models for finetune. There a guides for that e.g. https://www.youtube.com/watch?v=rEbpKxZbvIo&ab_channel=Roboflow image

cnnAndBn commented 3 years ago

hi,do you have the backbone weight pretrained in imagenet if I don't choose to use the released model trained in coco?

SashaAlderson commented 3 years ago

hi,do you have the backbone weight pretrained in imagenet if I don't choose to use the released model trained in coco?

No, i don’t have. I believe that this model will do, cause it’s trained to recognize different patterns and got a lot of data(from every image you’re getting a lot of boxes to train on). So i don’t see the problem here with Coco dataset. Correct me if i’m mistaken.