VIP: Visual-guided Prompt Evolution for Efficient Dense Vision-Language Inference

TL;DR

This paper introduces VIP, a novel method that enhances dense vision-language inference by evolving prompts with visual guidance, surpassing existing methods in accuracy and efficiency.

Contribution

VIP leverages visual-guided prompt evolution and a saliency-aware aggregation to improve semantic expressiveness and dense prediction quality in vision-language models.

Findings

VIP outperforms leading methods by 1.4%-8.4% in average mIoU.

VIP generalizes well across diverse challenging domains.

VIP requires minimal inference time and memory overhead.

Abstract

Pursuing training-free open-vocabulary semantic segmentation in an efficient and generalizable manner remains challenging due to the deep-seated spatial bias in CLIP. To overcome the limitations of existing solutions, this work moves beyond the CLIP-based paradigm and harnesses the recent spatially-aware dino$.$txt framework to facilitate more efficient and high-quality dense prediction. While dino$.$txt exhibits robust spatial awareness, we find that the semantic ambiguity of text queries gives rise to severe mismatch within its dense cross-modal interactions. To address this, we introduce Visual-guided Prompt evolution (VIP) to rectify the semantic expressiveness of text queries in dino$.$txt, unleashing its potential for fine-grained object perception. Towards this end, VIP integrates alias expansion with a visual-guided distillation mechanism to mine valuable semantic cues, which are robustly aggregated in a saliency-aware manner to yield a high-fidelity prediction. Extensive evaluations demonstrate that VIP: 1. surpasses the top-leading methods by 1.4%-8.4% average mIoU, 2. generalizes well to diverse challenging domains, and 3. requires marginal inference time and memory overhead.