Table 1.
Proposal results on VEDAI dataset. We evaluated 10 folds in VEDAI, where results are the mean values of 10 folds. The three metrics used were: average recall (AR), recall at IoU of 0.7, and average best overlap (ABO), which are recorded for three budgets of proposals: 500, 1000, and 2000. Values in bold indicate the best performances.
| Method | # Proposals = 500 | # Proposals = 1000 | # Proposals = 2000 | Time (sec) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AR | 70%-Recall | ABO | AR | 70%-Recall | ABO | AR | 70%-Recall | ABO | ||
| BING | 24.7 | 12.2 | 59.7 | 26.3 | 13.1 | 60.7 | 27.2 | 13.4 | 61.3 | 0.07 |
| SS | 15.7 | 15.0 | 43.4 | 26.1 | 27.9 | 56.6 | 38.4 | 44.8 | 66.6 | 19 |
| EB | 43.2 | 56.3 | 67.0 | 48.9 | 65.0 | 71.5 | 53.6 | 72.9 | 74.7 | 0.3 |
| EBNR-L | 48.1 | 62.8 | 72.5 | 53.3 | 72.2 | 75.3 | 57.0 | 78.6 | 77.0 | 0.36 |
| EBNR-G | 52.1 | 70.2 | 73.9 | 55.9 | 76.6 | 76.0 | 58.8 | 81.9 | 77.4 | 0.36 |