generalcritiquebearishPurely data-driven models for cross-modality image translation can produce visually plausible outputs that are inconsistent with optical image formationComputer Vision28 Jul 2026http://arxiv.org/abs/2607.21190v1