search

WongKinYiu / ScaledYOLOv4

Scaled-YOLOv4: Scaling Cross Stage Partial Network
GNU General Public License v3.0
2.02k stars 575 forks source link

Transfer learning with Pytorch codebase #115

Open matthai opened 3 years ago

matthai commented 3 years ago

Is there an example (or any hints/instructions) on how to do transfer learning with the pytorch code provided in this repo?

I understand that using darknet/partial works, but I'm interested in a pytorch example.

I'm a pytorch newbie, so if the answer is "just do it the usual pytorch way" please let me know, I will read up on it, although any hints will be appreciated!

WongKinYiu commented 3 years ago

if you use yolov4-csp, you can use darknet/partial to get coco pretrained weights for both darknet and pytorch. if you use yolov4-large, replace argument --weights '' to --weights pretrained.pt in your training command.

LukeAI commented 3 years ago

https://github.com/WongKinYiu/ScaledYOLOv4/issues/125

LukeAI commented 3 years ago

@matthai did you work this out in the end?

LukeAI commented 3 years ago

@WongKinYiu please could you show how to get partial pretrained weights for pytorch?

williamhyin commented 3 years ago

@WongKinYiu please could you show how to get partial pretrained weights for pytorch?

Hi Have you solved this problem. Start training yolov4-csp from scratch, and get *.pt results? I am really need solve this problem. Hope to get your help.