yolo4 custom object detection deep compression

[samohadid]

I want to thank you for your amazing work.
from a while ago I read your work, checked all issues, and the pull requests and I'm still confused. as mentioned In the title I'm trying to deep compress yolo4 weights file for a custom object.
After I read the paper "deep compression: compressing deep neural networks with pruning, trained quantization and Huffman coding" . I started searching for a reliable implementation. I Found that your work is the most famous and have covered many application.
The authors of the paper said that after pruning the weights file they train the network and re prune if needed.
I want to deep compress yolo4 custom object detection.
I found that you implemented all the pyimage models natively implemented in Pytorch (rcnn,faster rcnn...). I found an implantation for yolo4 in Pytorch yet I really think that avoiding to natively add darknet to this repository is the wisest approach.
I only need to compress the weight file. since darknet is vast and complicated Implementing or merging the two repositories is not an efficient approach and I'm really concerned about Pytorch not supporting open cv about the accuracy of yolo4 Pytorch.

• please confirm that this an implementation of the paper mentioned above.
• And a detailed step by step instructions for either adding yolo4 to that repository or to deep compress a custom stand alone way to only compress any weight file format(doesn't need to be yolo we can transform form yolo to other formats and vise versa) and retrain the network after compression natively in the original framework .