Can we say that Compression is done by down sampling?

Not necessarily, I think, since a compression algorithm may modify the points positions, not just take a subset of the vertices of a mesh.