Abstract: Current camouflaged object detection (COD) methods rely heavily on large-scale datasets with pixel-level annotations. We propose a semi-supervised iterative learning network (SILNet) to ...