Multilabel classification using ViT (based on facebookresearch dino repo)
python classify_single_image.py --image_path /PATH/TO/IMAGE --checkpoint /PATH/TO/MODEL_CHECKPOINT
DataLoader is created using torchvision.utils.ImageFolder, so images of each class should be in separate folders whose names are the class names.
--n_classes argument can be used to set the number of labels.