Enhancing JEPAs with Spatial Conditioning: Robust and Efficient Representation Learning