FusionSeg: Learning to combine motion and appearance for fully automatic segmentation of generic objects in videos Supplementary material

Table 1 shows the per video results for the 50 videos from the DAVIS dataset (referred in Table 1 of the main paper). We compare with several semi-supervised and fully automatic baselines. Our method outperforms the per-video best fully automatic and best semi-supervised baseline in 25 out of 50 videos. Table 2 shows the per video results for the 14 videos from the Segtrack-v2 dataset (referred in Table 3 of the main paper). Our method outperforms the per-video best fully automatic method in 5 out of 14 cases. Our method also outperforms the semi-supervised HVS [1] method in 8 out of 14 cases.