Table 1 Compare with the state-of-the-art methods on the dataset used in this paper. In the second box, we list the methods based on speedplus62 or other satellite image datasets, including the works by1,2,17,18,32,33. In the third box, we include the methods for human pose estimation, such as5,6,8. Best are highlighted and second are underlined.
From: SU-Net: pose estimation network for non-cooperative spacecraft on-orbit
Methods | MinAE | MaxAE | MeanAE | Std |
|---|---|---|---|---|
2021 Yang(AlexNet)33 | \(0.908^{\circ }\) | \(1.424^{\circ }\) | \(0.991^{\circ }\) | \(0.046^{\circ }\) |
2021 Yang(ResNet)33 | \(0.534^{\circ }\) | \(0.884^{\circ }\) | \(0.623^{\circ }\) | \(0.037^{\circ }\) |
2020 Lorenzo18 | \(0.942^{\circ }\) | \(1.426^{\circ }\) | \(1.023^{\circ }\) | \(0.052^{\circ }\) |
2020 Alexei32 | \(0.914^{\circ }\) | \(1.343^{\circ }\) | \(0.980^{\circ }\) | \(0.041^{\circ }\) |
2020 Shubham2 | \(0.897^{\circ }\) | \(1.222^{\circ }\) | \(0.959^{\circ }\) | \(0.043^{\circ }\) |
2020 Xu17 | \(1.803^{\circ }\) | \(4.051^{\circ }\) | \(2.161^{\circ }\) | \(0.640^{\circ }\) |
2022 Juan1 | \(\underline{0.303^{\circ }}\) | \(\underline{0.489^{\circ }}\) | \(\underline{0.401^{\circ }}\) | \(0.051^{\circ }\) |
2019 HRNet5 | \(0.583^{\circ }\) | \(0.699^{\circ }\) | \(0.661^{\circ }\) | \(0.034^{\circ }\) |
2021 MIPNet6 | \(0.576^{\circ }\) | \(0.661^{\circ }\) | \(0.646^{\circ }\) | \(\underline{0.031^{\circ }}\) |
2022 AggPose8 | \(0.551^{\circ }\) | \(0.639^{\circ }\) | \(0.625^{\circ }\) | \({{\bf 0.024}^{\circ }}\) |
SU-Net (our) | \({{\bf 0.128}^{\circ }}\) | \({{\bf 0.432}^{\circ }}\) | \({{\bf 0.282}^{\circ }}\) | \(0.063^{\circ }\) |
