Abstract
Zero-shot classification needs efficient label retrieval and fine-grained visual reasoning, yet discriminative and generative vision-language models fail in complementary this http URL CLIP's top-1 prediction is wrong, the correct label often remains in its top-$K$ shortlist, making disambiguation rather than recall the key this http URL generative models, however, are hindered by large label spaces and unconstrained this http URL complementarity motivates separating broad candidate retrieval from fine-grained, image-grounded this http URL propose G2D, a training-free framework that uses a generative VLM to verify CLIP-retrieved candidates against the this http URL names and CLIP probabilities provide a structured prior for resolving visually similar this http URL confidence routing, entropy-adaptive candidate sizing, and trie-constrained decoding focus generative reasoning on uncertain samples and ensure one valid output for each input at test this http URL eight benchmarks, G2D achieves 68.85% average accuracy, versus 59.35% for CLIP and 63.11% for the standalone this http URL seven generator configurations, candidate-set verification improves average accuracy by 1.08--27.42 percentage points.G2D also transfers to DCLIP, WaffleCLIP, and CuPL, supporting a practical interface between discriminative proposal and generative visual reasoning. Code: this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2608.26744
