MOCHA: Multi-modal Objects-aware Cross-arcHitecture AlignmentPublished in: Proceedings of the European Conference on Computer Vision (ECCV) Paper