What's the difference between yolov3.weights and darknet53.conv.74?

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)

MIT License

7.14k stars 3.45k forks source link

What's the difference between yolov3.weights and darknet53.conv.74? #521

Open YZAlex opened 5 years ago

YZAlex commented 5 years ago

I'm confused that which pre-trained weights file should I use when train a customized dataset, yolov3.weights or darknet.53.conv.74? Can darknet53.conv.74 act as weights file in training customized base model and tiny model? Thanks a lot~~

Ashwin-Ramesh2607 commented 4 years ago

You have to use darknet53.conv.74 This is because the yolov3.weights is tuned to detect 80 classes, and the last few layers and are aimed at capturing features for those 80 classes. Using the darknet53.conv.74 is therefore the correct option for retraining your model with a custom dataset.