Direct Segmentation without Logits Optimization for Training-Free Open-Vocabulary Semantic Segmentation

TL;DR

This paper introduces a training-free method for open-vocabulary semantic segmentation that directly derives semantic maps from distribution discrepancy without iterative training or attention modulation.

Contribution

It proposes an analytic solution to the distribution discrepancy for segmentation, eliminating the need for logits optimization and training.

Findings

Achieves state-of-the-art performance on eight benchmark datasets.

Eliminates the need for iterative training and model-specific attention modulation.

Provides a direct, analytic approach to semantic segmentation.

Abstract

Open-vocabulary semantic segmentation (OVSS) aims to segment arbitrary category regions in images using open-vocabulary prompts, necessitating that existing methods possess pixel-level vision-language alignment capability. Typically, this capability involves computing the cosine similarity, \ie, logits, between visual and linguistic features, and minimizing the distribution discrepancy between the logits and the ground truth (GT) to generate optimal logits that are subsequently used to construct segmentation maps, yet it depends on time-consuming iterative training or model-specific attention modulation. In this work, we propose a more direct approach that eschews the logits-optimization process by directly deriving an analytic solution for the segmentation map. We posit a key hypothesis: the distribution discrepancy encodes semantic information; specifically, this discrepancy exhibits consistency across patches belonging to the same category but inconsistency across different categories. Based on this hypothesis, we directly utilize the analytic solution of this distribution discrepancy as the semantic maps. In other words, we reformulate the optimization of the distribution discrepancy as deriving its analytic solution, thereby eliminating time-consuming iterative training, freeing us from model-specific attention modulation, and achieving state-of-the-art performance on eight benchmark datasets.