Mix and match networks enable image translation without paired data.
problem Image translation between domains or modalities with no direct paired data.
method Multiple encoders and decoders aligned for test-time composition.
result Model outperforms baselines in cross-modal image translation.