Keras-yolo3: Is it possible to use resnet backbone in Yolo v3?

[dmadhitama]

Is there any possibilities to use for instance resnet101 in the Yolo v3? If it is, any resource to find the .cfg and .weights file? Because when I referred to https://pjreddie.com/darknet/yolo/ there was not any resources for resnet backbone.

And I'm just curious, what is the backbone that being used in YOLOv3-320 model provided in pjreddie site?

[AlbertoSabater]

The backbone used in Yolo v3 is called Darknet-53.
To change the backbone, you should this function to fit the resnet101 architecture and later load its weights.

[dmadhitama]

The backbone used in Yolo v3 is called Darknet-53.
To change the backbone, you should this function to fit the resnet101 architecture and later load its weights.

How can I modify this function so I can change the backbone? What function do I have to modify? Or do I have to make my own function to fit the resnet101?

I'm also wondering because I just found the .cfg link in the pjreddie site here, can I use this .cfg and .weights files and then convert them to create new .h5 file contained resnet101 model, instead modify the model.py?

[AlbertoSabater]

You have to make your own function to fit the resnet101 architecture.
The .h5 file is just to load the weights into an already built NN.

[dmadhitama]

You have to make your own function to fit the resnet101 architecture.
The .h5 file is just to load the weights into an already built NN.

Alright then. Thank you for the answers!

[abdelDebug]

@dmadhitama Did you find please a solution to use resnet backbone in Yolo v3 ?