Multi-Label Test-Time Adaptation with Bayesian Conditional Priors

Researchers have developed Bayesian Conditional Priors (BCP) Estimation, a gradient-free test-time adaptation method that improves multi-label recognition with frozen Vision-Language Models under distribution shift. The approach corrects label co-occurrence errors by injecting label dependency through anchor-conditioned Bayesian refinement, using lightweight second-order statistics from unlabeled test data. On standard benchmarks, BCP boosted average mAP from 57.31 to 69.22 with RN50 and from 62.61 to 71.79 with ViT-B/16, outperforming strong test-time adaptation baselines.

arXiv:2606.12925v1 Announce Type: new Abstract: Multi-label recognition with frozen Vision-Language Models VLMs is brittle under distribution shift: standard zero-shot inference scores labels independently, ignoring co-occurrence structure and producing incoherent label sets where dominant concepts suppress weaker but compatible labels. We introduce Bayesian Conditional Priors BCP Estimation, a gradient-free test-time adaptation method that injects label dependency without tuning the backbone. BCP views zero-shot logits as a proxy for marginal posteriors under a fixed image-text likelihood and attributes shift-induced errors mainly to a mismatched label prior. For each test image, it selects a high-confidence anchor label and applies an anchor-conditioned Bayesian refinement. This update is closed-form in logit space and admits a pointwise mutual information PMI interpretation, explicitly promoting compatible labels and suppressing incompatible ones. BCP operates without target annotations by estimating anchor-conditioned priors online from the unlabeled test stream via lightweight second-order co-occurrence statistics, adding negligible overhead beyond a single forward pass. Across standard multi-label benchmarks and multiple CLIP backbones, BCP consistently outperforms strong TTA baselines, e.g., improving RN50 average mAP from 57.31 to 69.22 and ViT-B/16 from 62.61 to 71.79.