Difference between detection results.

[mariusfm54]

I recently wanted to compare darknet YOLOv3 inference results and ultralytics YOLOv3 (https://github.com/ultralytics/yolov3)
I converted the darknet weights .weights into PyTorch weights .pt using ultralytics archived repository and run detection using both darknet and ultralytics repo on the same dataset. I found out darknet results are more accurate.

I am thus wondering why is darknet more accurate: how does darknet perform detection?
For the PyTorch version, I used a YOLOv3 exclusive repository while darknet implements YOLOv4. and you don't need to precise what version of YOLO to use when running detection with darknet. So, could darknet run detection using YOLOv4 with YOLOv3 weights which makes results more accurate? What other reasons could cause darknet to be better?

Thank you for your answers.