It's interesting to see we can upsample via interpolation. However, how does the cow on the right have details than the left when it is obtained just via interpolation without prior knowledge about the object? My understanding is that such interpolation from a low-resolution mesh only increases the number of mesh but does not give additional fine details. Am I missing anything?
@huangwl18 I believe the subdivision of mesh involves approximating back to the original shape, and with the greater amount of mesh sampling we can better capture the shape details and even continuity this way.
@chelseayer so is it like supersampling in a way? where we sample "subpizels" in order to capture finer details/frequencies about the image itself.
Unlike supersampling, we don't have access to a higher quality "ground truth" mesh to sample new details from. Instead, we use algorithms like loop subdivision (https://cs184.eecs.berkeley.edu/sp20/lecture/8-41/meshes-and-geometry-processing) to interpolate new points from the points that are already present
@FLinesse good comment