1 paper
Oh-Hun Kwon, Eduard Zell
Several leading methods on public benchmarks for depth-from-stereo rely on memory-demanding 4D cost volumes and computationally intensive 3D convolutions for feature matching. We s…