Fig. 1. Point cloud reconstruction

Fig. 2. Pipeline of generating mask for training set

Fig. 3. Flowchart of the pose estimation algorithm

Fig. 4. Structure of neural network

Fig. 5. Selected 11 landmarks

Fig. 6. Transformation between satellite body reference and camera reference

Fig. 7. Accuracy of landmark localization

Fig. 8. Randomly selected images with the predicted segmentation of the satellite and the landmarks

Fig. 9. Same test images in Fig.8 with the predicted poses shown as the satellite coordinates