Yolov3-tiny-XNOR and Yolov3-Tiny Hybrid

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )

http://pjreddie.com/darknet/

Other

21.63k stars 7.95k forks source link

Yolov3-tiny-XNOR and Yolov3-Tiny Hybrid #2925

Open AbanoubMamdouh opened 5 years ago

AbanoubMamdouh commented 5 years ago

Hello, it's more of a question than an issue, is it possible to train my Yolov3-tiny network to have a hybrid configuration between the standard and the XNOR implementation of yolov3-tiny? For example, is it okay in the cfg file to set the XNOR=1 in some layers and not in some others? From what I read, there should be no problem, anybody has another opinion? Also, if this is possible, which weights file should I use? The XNOR one or the standard one? Thanks in advance

AlexeyAB commented 5 years ago

@AbanoubMamdouh Hi,

Yes, you can use hybrid cfg-file. The default yolov3-tiny-xnor is Hybrid cfg, because you should use FP32 for the 1st and last conv-layers.

You can set xnor=1 for any convolutional layers.