Table 1 Performance comparison of our method with the state-of-the-art methods on the NEU-Seg dataset.
From: Hyperbolic geometry enhanced feature filtering network for industrial anomaly detection
Method | Background | Inclusion | Patch | Scratch | mIoU | Parameters | FLOPs |
|---|---|---|---|---|---|---|---|
FCN33 | 96.89 | 67.49 | 82.40 | 72.55 | 79.83 | 9.835M | 2.925G |
PSPNet34 | 97.32 | 71.93 | 85.58 | 75.24 | 82.52 | 44.54M | 27.302G |
DeepLabV3+35 | 97.05 | 71.15 | 84.94 | 78.72 | 82.96 | 56.49M | 26.932G |
EMANet36 | 97.33 | 72.14 | 85.96 | 72.97 | 82.25 | 56.17M | 25.868G |
FPN37 | 97.01 | 70.72 | 85.81 | 78.12 | 82.91 | 28.496M | 7.237G |
ICNet38 | 97.07 | 67.88 | 85.20 | 72.01 | 80.54 | 24.847M | 1.379G |
CGNet39 | 96.77 | 66.64 | 83.38 | 70.80 | 79.40 | 0.496M | 0.529G |
STDC240 | 97.42 | 69.61 | 85.38 | 72.39 | 81.13 | 12.306M | 1.911G |
STDC140 | 97.23 | 71.14 | 85.98 | 72.33 | 81.82 | 8.275M | 1.365G |
BiSeNetV141 | 97.00 | 67.24 | 84.61 | 73.01 | 80.46 | 13.276M | 2.384G |
BiSeNetV242 | 97.27 | 71.14 | 85.95 | 73.55 | 81.98 | 3.359M | 1.891G |
Fast-SCNN43 | 97.11 | 69.12 | 85.11 | 71.66 | 80.75 | 1.4M | 0.155G |
DDRNet44 | 97.24 | 68.58 | 86.41 | 72.85 | 81.27 | 6.023M | 0.74G |
FDSNet*45 | – | – | – | – | 78.80 | 0.97M | – |
RTFormer46 | 97.35 | 67.95 | 85.32 | 73.54 | 81.04 | 5.21M | 0.67G |
Trans4Trans47 | 97.14 | 68.51 | 83.49 | 72.97 | 80.52 | 13.11M | 1.52G |
PIDNet48 | 97.10 | 71.27 | 84.70 | 72.06 | 81.28 | 7.721M | 0.966G |
SegFormer49 | 96.96 | 67.19 | 83.42 | 77.04 | 81.15 | 3.716M | 1.096G |
SeaFormer50 | 97.26 | 73.53 | 84.42 | 73.4 | 82.15 | 1.65M | 0.1G |
SCTNet51 | 97.14 | 71.79 | 83.92 | 73.72 | 81.65 | 29.72M | 0.58G |
ConvNext52 | 96.82 | 65.52 | 82.71 | 73.65 | 79.68 | 123.91M | 43.634G |
SegNext53 | 97.05 | 69.44 | 84.13 | 73.23 | 80.96 | 4.226M | 1.001G |
PoolFormer54 | 96.80 | 65.93 | 82.41 | 75.34 | 80.12 | 15.561M | 4.874G |
DDSNet55 | 97.74 | 76.87 | 88.91 | 76.96 | 85.12 | 9.747M | 0.97G |
Ours | 98.10 | 79.45 | 90.83 | 79.69 | 87.00 | 11.352M | 1.113G |
