training-detr
training-detr copied to clipboard
looking forward to your notebook
Hi @lessw2020, I am very looking forward to your colab! Meanwhile, shouldn't you freeze all the parameters except for the head before training? And yes, showing how to obtain a custom dataset in coco format would be great!!
Thanks a lot!
Hi @tazu786 - I had no issue simply training the whole thing rather than doing the freezing of the head. However, it's a valid point that we should test freezing and training with just the new head and compare that vs training the whole thing. I've added a custom class that does the remapping to allow for coco format custom datasets as well now.