From Fig. 8, we can see that based on the four bias compensation models, the results of detection of the collapsed buildings are generally consistent. Moreover, the results based on the affine and second-order models seem a little better, particularly at the points from 25 to 36 (the values are much closer to zero, showing that the buildings have not collapsed in the earthquake). Therefore, accounting for both stability and accuracy, the affine bias-compensation model was adopted for 3D accuracy improvement in the study.