search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.71k stars 7.96k forks source link

yolov4-tiny convert to caffe #6583

Open lesterlee89 opened 4 years ago

lesterlee89 commented 4 years ago

Hi guys: yolov4-tiny has the route: [route] layers=-1 groups=2 group_id=1

How to convert yolov4-tiny to caffe without change the caffe layers, or which layer could achieve the function thanks

WongKinYiu commented 4 years ago

i think you can use slice layer instead.

lesterlee89 commented 4 years ago

thanks for your reply, if use the slice layer, there will be half channel to be left without use in an resblock, and the outputs of the network will be increase three more.

Hwijune commented 4 years ago

hi! @lesterlee89 Did you convert the route layer into a caffe?

lesterlee89 commented 4 years ago

no, i use the 1x1 conv to replace the route layer

alexanderfrey commented 3 years ago

no, i use the 1x1 conv to replace the route layer

Can you show an example on how ?

alexanderfrey commented 3 years ago

I think the idea is to throw away the first part of the feature map and only route the second part. Is that correct ? Maybe @AlexeyAB can provide some insights or link that would explain the logic behind the group routing ?

Thanks !

MuhammadAsadJaved commented 3 years ago

@alexanderfrey @lesterlee89 I am trying to convert Yolov4-tiny darknet to caffe. Can you please give some details about how I can convert it? Which project and changed did you made? Can you share your project?

Looking forward to hearing from you.

Hiwyl commented 3 years ago

@alexanderfrey @lesterlee89 Do you have some good solutions?Looking forward to hearing from you.