Experiment on effect of differently cropped shape on SSM

In order to achieve better validation result, I am trying to use the shape models that were cropped differently according to target object.
As the back of subject cannot been seen by both Kinects and the plastic torso is rigid and not as same as human torso, I tried to use shapes whose back have been cropped (shown in Figure 1).

Figure 1 Torso shape model (no head, legs, arms and back)

For real human subject, since the head and arms are required (no head nor arms for the plastic torso), I remain the head and part of arm for shape model training (shown in Figure 2).

Figure 2 Torso shape model (no legs, part arms and no back)