Let’s start with the product pile that was on site. As you may recall, I captured it with both a Leica C10 and with a DJI Phantom 3 Professional. At first, both data sets looked great. In order to eliminate as many variables as possible, I exported a point cloud from Pix4D and brought it into Leica Geosystems Cyclone for comparison with the C10 data. Since you can choose the density of point cloud when creating it in Pix4D, density issues would seem to be a moot point. However, I am of the opinion that point sampling from a mesh is like choosing the number of digits after zero when measuring: You should not sample at an interval that is greater than your per-pixel resolution as it gives the impression of higher accuracy than actually exists.
That being said, our Ground Sampling Distance was 1.2 in/pixel (3.05 cm/pixel) so we set our point cloud sampling spacing to match. The first thing that jumps out is how much denser the C10 data is. However, this has a much smaller effect on the overall accuracy then our next discovery.