ACV-Gate Framework for Efficient Visual Token Communication

ACV-Gate uses an adaptive candidate evaluation framework to reduce encoder-side computation in generative image communication. It approximates full-budget counterfactual evaluation by using a set-aware student to predict token rankings and a selective refinement mechanism to limit exact evaluations.