I was search most of the sources i think, but can not find how to do. I use KITTI dataset from pytorch training_set = torchvision.datasets.Kitti(root, train=True, download=True,transform=transform) with datasets.wrap_dataset_for_transforms_v2 so that wrapper makes 0:car 1:van 2:truck ... to 8:DontCare I think,
want to use retinanet and faster r-cnn, when I look up their source code there is warning as __ground__ is default 0 so add to your classes +1 but i don’t know is my dataset labels add. or my model try to 0:__ground__ and 0:car same time? i did not get error while training or draw it but classification of objects are terrible but bboxes are pretty good. could you help me? thanks in advance for even read.