left-right labelling

[mureva]

Hi,

I am experimenting with DensePose to see if it is useful for some data I'm processing. What I currently find however is that the body part labelling is often confused between left and right, particularly on the legs and especially on the feet. Worse is that it can often label half the leg as left, and half as right, or even have two left legs.

Is there one of the pre-trained models that is known to be better at resolving left-right, or a setting in the config file that can improve left-right, or a way of giving the network a prior on what is left and what is right, or even, for it to output multiple results ( as in, if this was left, these are the UVs and if it is right, these are the UVs, and this is my probability on this being left or right body part) ?