Every time I think I found a resource that explains how the automatic silhouette stuff works it turns out to be something else.
I thought this paper would go into it:
http://amd-dev.wpengine.netdna-cdn.com/wordpress/media/2012/10/Tatarchuk-ParallaxOcclusionMapping-FINAL_Print.pdf
but then at the end of the paper this is their solution to corners:
I’d appreciate if you find any resources that talk about that part specifically. There is an nvidia/speedtree tree material that may be useful here but it may rely on the 'fin’s they create. not sure yet.