We’re using YOLOv2 for industrial purposes, but loss is not convering / mAP is low.
We’re looking for someone with experience training YOLOv2 networks to help us with advice or hands-on training – happy to pay for their time!
Hi there.
Are you using original darknet or other derivatives for training? Even if you copy network structure and train on tf or pyroch results may vary from the original.
It was a while but I remember initial training from a new dataset needs some changes in the network in order to generalize (check the net config git history)
Be sure that you check your dataset format and bboxes format before jump into training. Usually people made mistakes during dataset preparation.