Hugh Hoppe, from Microsoft, wrote some fundamental progressive-mesh code that he showed at SIGGRAPH '97 using this model. Instead of the code having to choose from different model resolutions (like the original four bunny models), the algorithm would allow DirectX to reduce the vertex count based on the rendered size of the object (1 pixel only needs 1 vertex!) to save compute cycles.

I wonder if geometry compression is still a thing given the performance increases of GPUs. I haven't done 3D driver work since DirectX 5....