search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.6k stars 7.95k forks source link

Yolov3 Fine-tuning using yolov3.conv.81 #3072

Open ghost opened 5 years ago

ghost commented 5 years ago

Hi, In link, it is mentioned to add stopbackward = 1 to freeze the 81 layers of yolov3.weights. However there are only 75 convolutional layers in yolov3.cfg file. How is that possible?

Thanks,

AlexeyAB commented 5 years ago

Will be freezed 81 layers, just it will affect only on CONVOLUTIONAL, CRNN, RNN, LSTM, CONV-LSTM, Batch-norm layers.

ghost commented 5 years ago

I am still confused. When I look at yolov3.cfg file, I can see [convolutional] and [shortcut]. Could you explain more the meaning of layer here?

AlexeyAB commented 5 years ago

All [...] are layers except [net]

ghost commented 5 years ago

So in that case, we are freezing 75 layers in link not 81, right?

AlexeyAB commented 5 years ago

Yes.