propagation on the DAVIS dataset (Table 1), in comparison to a SOT A3 self-supervised method [49] and the ImageNet pre-trained representation