Area occlusion

I provided the RGB values, along with the target's mask and pointmap. However, the image has occlusion issues, causing the prediction to deviate from the actual values ​​in occluded areas. I'd like to know if the prediction model has a threshold setting. In other words, should the prediction of the mask containing the pointmap be accurate, and should the completion of the region be determined based on a threshold to determine whether to retain it?