Calibrate What You SHIP: Post-Selection Risk Control for Verifier-Guided Text-to-Image Generation

2026-08-21

AI Summary

本文针对验证器引导的文本到图像生成系统中候选与策略校准不匹配的问题，提出SHIP方法进行选择感知的保留校准，以控制实际发布输出的风险。

Abstract

Verifier-guided text-to-image systems increasingly use test-time search to select, refine, or stop among multiple candidates, yet release thresholds are often calibrated on individual images. This creates a candidate-to-policy calibration mismatch: search changes both which prompts receive an output and which candidate is released, so candidate-level risk control need not imply control of released-output risk. We formalize this estimand shift through prompt reweighting and within-prompt selection, and introduce SHIP, Selection-aware Held-out calibration of Inference Policies. SHIP runs or replays the complete deployed policy on held-out prompts, evaluates the image it actually releases using an independent target judge, and selects the most permissive threshold whose risk upper bound satisfies a prescribed budget. For replayable policies with a prespecified threshold grid, simultaneous confidence control provides finite-sample validity. Experiments across fixed, sequential, and adaptive T2I inference procedures show that policy-level calibration recovers lower-risk operating points while exposing policy-dependent tradeoffs among risk, coverage, and compute. On GenEval2 with FLUX at N=16, a pooled-candidate threshold yields released risk 0.310, whereas SHIP reduces it to 0.162. Across 200 cached-stream splits, the fixed-grid certificate has no target crossing. Reliable inference-time scaling therefore requires calibrating the output distribution induced by the complete deployed policy.

Problem

• Verifier-guided
• Text-to-Image Generation
• Post-Selection Risk Control
• Calibration Mismatch
• Inference Policies

Innovation

• SHIP
• calibration
• text-to-image generation
• risk control
• policy-level