A temporal repository for S2SiamFC.
The results of VOT2016 & VOT2018 are under ./results.
Note: The following results are based on existing experimental results (the first two rows in Table 3 in the paper) but in detail. Therefore, it should obey the rebuttal policy.
↑ denotes larger is better; ↓ denotes smaller is better
There are some some successful cases that can illustrate the contributions of anti-clutter weighting:
S2SiamFC with AC weighting | S2SiamFC without AC weighting | |
Video name | Lost Number ↓ | Lost Number ↓ |
ball2 | 1.000 | 2.000 |
birds1 | 3.000 | 4.000 |
blanket | 1.000 | 2.000 |
book | 3.000 | 5.000 |
butterfly | 1.000 | 3.000 |
conduction1 | 0.000 | 4.000 |
fish2 | 7.000 | 8.000 |
fish3 | 2.000 | 3.000 |
flamingo1 | 6.000 | 8.000 |
gymnastics1 | 7.000 | 9.000 |
gymnastics2 | 4.000 | 6.000 |
handball1 | 7.000 | 8.000 |
handball2 | 5.000 | 6.000 |
iceskater1 | 7.000 | 11.000 |
iceskater2 | 5.000 | 8.000 |
leaves | 4.000 | 5.000 |
nature | 5.000 | 6.000 |
traffic | 0.000 | 1.000 |
zebrafish1 | 17.000 | 19.000 |
iceskater1 | 5.000 | 11.000 |
road | 1.000 | 5.000 |
soccer1 | 9.000 | 10.000 |
VOT2018 | Accuracy ↑ | Robustness ↓ | EAO ↑ |
---|---|---|---|
S2SiamFC with AC weighting | 0.476 | 1.110 | 0.136 |
S2SiamFC without AC weighting | 0.472 | 1.232 | 0.130 |