search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.77k stars 7.96k forks source link

pre-tained weights for tiny YOLO #104

Open fishman2008 opened 7 years ago

fishman2008 commented 7 years ago

Hi AlexeyAB,

If I want to train a tiny yolo network, which cfg file should I use? tiny-yolo-voc.cfg, tiny-yolo.cfg or tiny.cfg? Besides, what will be the pre-trained weights from imagine, similar to the one used for darknet19_448.conv.23? Thanks.

AlexeyAB commented 7 years ago

@fishman2008 Hi, Use tiny-yolo-voc.cfg and darknet19_448.conv.23 to train Tiny-model. With all training steps as there: https://github.com/AlexeyAB/darknet#how-to-train-to-detect-your-custom-objects

fishman2008 commented 7 years ago

Thanks for your quick response, AlexeyAB. I have trained the model using the provided setting. But not sure if it is correct, just want to double check on that.

darknet19_448.conv.23 appears to achieve very good performance on ImageNet1K. The cfg directory also contain RESNET cfg. Would it be possible to train a RESNET model on ImageNet and then use it as a pre-trained weight. In this case, how should I modify the cfg file for detection? Thanks.