Table 6 Evaluation results across different target domains, illustrating performance under varying levels of domain shift.
From: Inspection of railway catenary systems using machine learning with domain knowledge integration
Test domain | Method | AP50 | Recall | Precision | F1 |
|---|---|---|---|---|---|
Basic elements | |||||
Osielec 1+2 (training) | Met.1 | 52.19 (3.78) | 67.48 (3.48) | 50.96 (3.05) | 54.80 (3.11) |
Met.3+D+M | 51.22 (4.17) | 66.79 (3.96) | 53.61 (4.13) | 56.94 (3.22) | |
Osielec 3 | Met.1 | 56.54 (5.39) | 70.34 (5.75) | 60.73 (2.93) | 64.11 (4.50) |
Met.3+D+M | 55.03 (3.78) | 67.84 (4.40) | 60.65 (4.49) | 63.25 (3.56) | |
Jeleśnia | Met.1 | 22.35 (3.80) | 29.63 (4.70) | 45.74 (14.65) | 30.22 (4.85) |
Met.3+D+M | 19.44 (2.78) | 24.44 (2.98) | 48.13 (14.17) | 28.56 (3.21) | |
Poronin | Met.1 | 59.14 (10.10) | 68.74 (10.58) | 61.83 (5.56) | 64.76 (7.11) |
Met.3+D+M | 63.01 (6.37) | 66.61 (5.60) | 83.13 (8.91) | 73.43 (4.96) | |
Small elements | |||||
Osielec 1+2 (training) | Met.1 | 23.27 (3.57) | 47.88 (6.98) | 20.48 (2.76) | 27.76 (3.04) |
Met.3+D+M | 66.72 (5.35) | 83.39 (3.50) | 41.40 (5.45) | 55.13 (5.59) | |
Osielec 3 | Met.1 | 16.42 (2.13) | 34.07 (4.20) | 27.32 (3.58) | 25.96 (3.68) |
Met.3+D+M | 57.92 (4.18) | 76.85 (4.56) | 52.51 (2.99) | 62.09 (3.17) | |
Jeleśnia | Met.1 | 7.91 (1.68) | 14.96 (2.76) | 20.50 (11.30) | 15.86 (2.75) |
Met.3+D+M | 19.52 (2.91) | 22.86 (3.27) | 64.11 (14.45) | 30.56 (2.93) | |
Poronin | Met.1 | 11.94 (2.61) | 21.50 (5.88) | 18.79 (2.44) | 19.24 (1.81) |
Met.3+D+M | 28.66 (4.43) | 29.25 (4.41) | 47.36 (1.63) | 35.99 (3.61) | |
