Hi!
I have pretrained yolov8 model from roboflow dataset and I want to use it with my custom dataset to increase its output that’s why I decided to use trasfer learning and at the end I want to use new model with deepstream.
BUT, I have some doubts how I can start which steps and any tutorial about it. i.e any hardware limitations about transfer learning, any colab notebook or anything else I am asking them because I deep dived TAO toolkit but its pretrained model didn’t fit my dataset that’s why I pass another transfer learning options to fit deepstream at the end.
I just thought that just taking weights and cfg of my pretrained model and other dataset ad train them bu there is freezing layer part and is it related to transfer learning or not is not clear for me.
Is there any easy to follow tutorial to see my o/p? I hope I described my issue, everything is unclear, need to create new roadmap for this.
EDIT:
Also in my roboflow pretrained model has different labels than my dataset. So in that case is there anyconfig file do I need to customize?
Also my dataset type will affect the model while transfer learning or do I need to be strict for yolo format dataset?
Also if dataset images are in different sizes? Does it affect accuracy?
Thanks!