Table 5 Comparison of benchmark evaluation results for component segmentation.
From: A Large-Scale Synthetic Benchmark Dataset for Non-Cooperative Space Target Perception
Method | mIOU | mACC | Flops(G) | Params(M) | backbone | Body | Solar panel | Antenna | Observation load | ||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Iou | Acc | Iou | Acc | Iou | Acc | Iou | Acc | ||||||
FCN | 77.7 | 83.2 | 58.0 | 47.1 | Resnet50 | 69.52 | 72.75 | 90.04 | 94.19 | 75.73 | 81.29 | 75.57 | 84.72 |
Unet | 80.3 | 85.8 | 203 | 29.0 | Resnet50 | 71.83 | 75.59 | 91.24 | 96.64 | 80.0 | 87.14 | 78.21 | 83.73 |
DeepLabV3+ | 83.0 | 88.7 | 177 | 41.2 | Resnet50 | 72.86 | 76.56 | 93.53 | 96.52 | 84.11 | 91.95 | 81.46 | 89.71 |
Gcnet | 81.4 | 86.2 | 198 | 47.3 | Resnet50 | 72.06 | 75.69 | 92.47 | 95.54 | 81.75 | 88.24 | 79.28 | 85.25 |
Fast-SCNN | 43.90 | 56.0 | 0.936 | 1.40 | — | 41.37 | 74.49 | 45.94 | 46.48 | 38.15 | 40.41 | 46.44 | 62.46 |
SETR | 71.2 | 79.7 | 367 | 309 | ViT-Large | 65.43 | 71.42 | 86.42 | 92.86 | 66.38 | 78.78 | 66.37 | 75.9 |
Segformer | 79.9 | 86.4 | 7.90 | 3.72 | Mit_b0 | 70.47 | 75.07 | 91.48 | 95.6 | 79.94 | 87.34 | 78.0 | 87.44 |
K-net | 82.4 | 87.9 | 273 | 78.9 | Resnet50 | 73.4 | 76.38 | 92.26 | 96.81 | 83.02 | 90.14 | 80.87 | 88.61 |
MaskFormer | 84.8 | 89.5 | 181 | 45.0 | Resnet50 | 74.5 | 77.33 | 93.44 | 96.61 | 86.69 | 93.48 | 84.48 | 90.72 |
Mask2Former | 88.1 | 91.0 | 226 | 44.0 | Resnet50 | 76.66 | 78.25 | 95.69 | 97.75 | 91.28 | 95.01 | 88.8 | 93.1 |
