Table 3 Performance comparison on the FSSD-12 dataset.
From: Hyperbolic geometry enhanced feature filtering network for industrial anomaly detection
Method | AM | La | Ld | Op | Os | Pa | Pk | Ri | Rp | Sc | Se | Ws | mIoU |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Fast-SCNN | 68.32 | 63.28 | 72.65 | 60.53 | 52.09 | 61.96 | 63.89 | 50.81 | 61.26 | 58.22 | 47.15 | 49.93 | 59.17 |
SegNext | 73.13 | 74.62 | 86.28 | 71.20 | 67.36 | 76.96 | 75.74 | 68.15 | 73.11 | 69.23 | 58.74 | 59.01 | 71.12 |
SCTNet | 74.93 | 71.24 | 85.41 | 73.37 | 65.22 | 74.40 | 78.28 | 64.83 | 72.69 | 67.07 | 59.34 | 58.21 | 70.42 |
RTFormer | 71.94 | 68.56 | 79.45 | 73.11 | 61.10 | 70.89 | 79.57 | 72.51 | 74.36 | 66.27 | 53.22 | 54.09 | 68.76 |
SeaFormer | 75.33 | 70.42 | 83.97 | 72.51 | 64.89 | 74.20 | 79.23 | 56.28 | 74.25 | 66.29 | 57.83 | 59.47 | 69.56 |
Trans4Trans | 72.85 | 69.32 | 81.28 | 70.12 | 63.45 | 71.24 | 75.56 | 62.43 | 71.8 | 65.37 | 54.32 | 56.11 | 67.82 |
ConvNext | 73.97 | 71.06 | 84.56 | 75.18 | 65.20 | 73.65 | 80.37 | 66.24 | 74.93 | 69.05 | 58.96 | 57.41 | 70.88 |
DDSNet | 75.39 | 74.27 | 87.70 | 76.69 | 68.08 | 76.11 | 83.58 | 70.65 | 77.97 | 72.77 | 62.01 | 60.79 | 73.83 |
Ours | 81.72 | 77.80 | 90.81 | 81.66 | 70.72 | 81.47 | 76.22 | 73.85 | 80.64 | 77.17 | 67.53 | 64.91 | 77.04 |
