I know I said we should leave this alone because it looks like PDS has nailed this, but actually I realized how easy it would be to do a good job now that we have a good rendering capability. Therefore I put together pg_rendernav, which looks for the best correlation between an image and rendered scene. It is very simple, and it works extremely well. I found an image that fools it, but I think the cause is that the search wraps the image offsets, which is not optimal. The nice thing is that this procedure will naturally improve as we add maps and improve the rendering.
From @jnspitale on February 22, 2017 22:2
I know I said we should leave this alone because it looks like PDS has nailed this, but actually I realized how easy it would be to do a good job now that we have a good rendering capability. Therefore I put together pg_rendernav, which looks for the best correlation between an image and rendered scene. It is very simple, and it works extremely well. I found an image that fools it, but I think the cause is that the search wraps the image offsets, which is not optimal. The nice thing is that this procedure will naturally improve as we add maps and improve the rendering.
Copied from original issue: ppenteado/ominas#5