Table 1 Comparison of results on the UAVid test set with other methods. The bold column indicates the best results.
From: Large language model-driven knowledge graph reasoning for enhanced semantic segmentation
Method | Building | Road | Tree | Vegetation | MovingCar | StaticCar | Human | Clutter | mIoU | Params(M) | GFlops |
|---|---|---|---|---|---|---|---|---|---|---|---|
UnetFormer56 | 87.40 | 81.50 | 80.20 | 63.50 | 73.60 | 56.40 | 31.00 | 68.40 | 67.80 | 11.7 | 56.9 |
DecoupleNet D257 | 85.40 | 80.60 | 78.80 | 62.10 | 74.10 | 49.70 | 30.80 | 65.10 | 65.80 | 6.8 | 32.1 |
SegFormer58 | 86.30 | 80.10 | 79.60 | 62.30 | 72.50 | 52.50 | 28.50 | 66.60 | 66.00 | 13.7 | 63.3 |
DDRNet (baseline)37 | 91.64 | 84.17 | 78.82 | 71.95 | 72.09 | 67.65 | 27.14 | 70.64 | 70.51 | 5.73 | 4.91 |
Ours | 91.79 | 83.62 | 78.56 | 71.84 | 74.15 | 69.50 | 27.97 | 70.08 | 70.94 | 5.79 | 4.93 |
