SingleClassRL

Implementation of weak single object segmentation from paper "Regularized Loss for Weakly Supervised Single Class Semantic Segmentation", ECCV2020. PDF

Main Files

• train_with_anneal.py: use for training first in annealing stage, then in normal stage
• train_with_transfer.py: use from training with weight transfer from another dataset, models that can be used for weight transfer are in directory 'trained_models'

OxfodPet dataset

Download OxfordPet from (https://www.robots.ox.ac.uk/~vgg/data/pets/) Files in 'SingleClassRL\data\Oxford_iit_pet\annotations' should be placed in the 'annotation' directory of OxfordPet dataset