Table 2 The ablation study results in terms of the ADD(-S) metric on the Occlusion Linemod dataset.
From: Enhancing object pose estimation for RGB images in cluttered scenes
Object | w/o MHSA | w/o FPN | w/o IR | (4 steps) IR | Resnet | Efficientnet | Densenet (Baseline) | \(\phi =2\) |
|---|---|---|---|---|---|---|---|---|
Ape | 54.95 | 55.35 | 51.37 | 60.13 | 55.53 | 53.23 | 59.80 | 66.46 |
Can | 89.37 | 91.32 | 92.41 | 93.25 | 92.68 | 93.95 | 93.17 | 94.73 |
Cat | 58.77 | 52.63 | 50.35 | 64.37 | 60.85 | 58.77 | 66.80 | 65.41 |
Driller | 94.08 | 95.15 | 96.22 | 96.71 | 95.83 | 95.34 | 96.51 | 97.09 |
Duck | 63.96 | 59.06 | 66.67 | 73.44 | 66.77 | 68.02 | 72.40 | 78.85 |
Eggbox | 92.66 | 93.26 | 92.66 | 95.15 | 92.96 | 92.25 | 94.87 | 95.07 |
Glue | 89.62 | 86.47 | 88.96 | 85.67 | 86.47 | 87.78 | 86.47 | 90.28 |
Holepuncher | 82.62 | 80.57 | 82.84 | 85.02 | 80.47 | 84.42 | 84.62 | 87.18 |
Average | 78.18 | 76.73 | 77.69 | 81.72 | 78.92 | 79.22 | 81.84 | 84.38 |
