(reported by Willi Koller)
... or make it toggle-able in the model warper.
Willi reported that the model warper produces different-looking meshes from the mesh warper when given the same landmark data. I investigated the issue and it turns out that:
- The fact that the mesh warper has different transform defaults (scale+rotate+translate+warp vs. just scale+warp) makes it confusing to compare the outputs from both processes.
- The model warper, by default, doesn't recompute the warped mesh's normals (legacy behavior), whereas the TPS mesh warping step does. This makes the resulting meshes look different, which makes validation and sanity checks harder.