Table 5.
Experimental results on both GrapeCS-ML and our internal dataset based on the number of bunches present in the images. After each mAP value, in brackets, the number of examined images is shown.
| Dataset Name | mAP (Total Number of Images) | |||||
|---|---|---|---|---|---|---|
| 1 bunch | 2 bunches | 3 bunches | 4 bunches | 5 bunches | 6 bunches | |
| Validation (Set 2) | 98.85% (369) | 82.61% (126) | 57.22% (10) | |||
| Test (Set 3, 4, 5) | 99.75% (395) | 65.41% (73) | 72.59% (15) | 51.72% (8) | 60.00% (2) | 64.63% (2) |
| Set 3 | 100.00% (195) | 72.22% (9) | ||||
| Set 4 | 99.45% (181) | 57.70% (53) | 65.28% (8) | |||
| Set 5 | 100.00% (19) | 96.97% (11) | 80.95% (7) | 51.72% (8) | 60.00% (2) | 64.63% (2) |
| Internal Dataset | 96.79% (218) | 85.39% (166) | 76.11% (46) | 80.89% (17) | 99.17% (4) | |