Open gd1925 opened 2 years ago
If your custom dataset is from natural images and/or the objects in your dataset are similar to objects in COCO dataset then you can use transfer learning, else you have to train from scratch
Thank you so much for your reply.
Did you do it? how are the results? was transfer learning succesfull or you trained it from scratch?
Thanks in advance.
Hi all,
I wish to understand a concept in training. I want to add some objects for detections on top of the objects that the model already detects. (The model for example which is trained on 80 classes and I wish to add maybe 2-3 more classes to it). So is there any way to do it without having to train it from the start? Can transfer learning be useful in this case? Thank you so much.