I am trying to implement mesh smoothing algorithm mentioned here
But I got the problem with the delta of the vertex position (formula 13). The mean curvature flow is divided by the sum of the triangle areas. If the triangle areas are small, the mean curvature flow and mean curvature become too big values (mean curvature is estimated as the half of the magnitude of the mean curvature flow) and the point are moved far away from their positions. To make it do not crash my mesh I have to set lambda to very small value. Is it possible to make the algorithm do not depend on the scale of the model(distances between points) by normalizing as it said(Section 5.5) here? Also I do not see any visible changes in mesh even after 200 iterations(Stanford bunny). Is it ok?
Update: After a bit more research I have found this article http:// copyme.github.io/flower/mean-curvature/ (sorry, I cannot post more than 2 links) The do not divide by area. I tried it for me and it worked! Why? Have I divide by area on the meshes with big enough distances between points or do not divide at all? And how in the case of not dividing by triangle area would be the gaussian curvature? Maybe something like this: $$K = 2*\pi-\sum \phi$$, where $\phi$ is on of the angles surrounding current vertex