免调参的通用监督语义分割
Tuning-free Universally-Supervised Semantic Segmentation
AI总结：
提出基于CLIP分类SAM掩码的免调参语义分割框架，通过判别偏差对齐和全局-局部一致分类器，在多种监督下达到SOTA性能。
摘要：
本工作提出了一种基于CLIP对SAM掩码进行分类的免调参语义分割框架，该框架普遍适用于各种类型的监督。首先，我们利用CLIP的零样本分类能力生成伪标签或执行开放词汇分割。然而，掩码与CLIP文本嵌入之间的错位导致次优结果。为解决此问题，我们提出判别偏差对齐的CLIP，以紧密对齐掩码和文本嵌入，提供无额外开销的性能提升。随后，我们构建一个全局-局部一致分类器来对SAM掩码进行分类，该分类器揭示了DBA-CLIP产生的高质量嵌入的内在结构，并展示了对噪声伪标签的鲁棒性。大量实验验证了我们方法的效率和有效性，我们在各种数据集和监督类型上达到了最先进（SOTA）或具有竞争力的性能。
英文摘要：
This work presents a tuning-free semantic segmentation framework based on classifying SAM masks by CLIP, which is universally applicable to various types of supervision. Initially, we utilize CLIP's zero-shot classification ability to generate pseudo-labels or perform open-vocabulary segmentation. However, the misalignment between mask and CLIP text embeddings leads to suboptimal results. To address this issue, we propose discrimination-bias aligned CLIP to closely align mask and text embedding, offering an overhead-free performance gain. We then construct a global-local consistent classifier to classify SAM masks, which reveals the intrinsic structure of high-quality embeddings produced by DBA-CLIP and demonstrates robustness against noisy pseudo-labels. Extensive experiments validate the efficiency and effectiveness of our method, and we achieve state-of-the-art (SOTA) or competitive performance across various datasets and supervision types.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
