MSGCoOp: Multiple Semantic-Guided Context Optimization for Few-Shot Learning

Vision-language pre-trained models (VLMs) such as CLIP have demonstrated remarkable zero-shot generalization, and prompt learning has emerged as an efficient alternative to full fine-tuning. However, existing prompt learning methods often rely on a single prompt per class, failing to capture diverse semantic aspects, and use limited semantic anchors, hindering generalization to novel categories. In this paper, we propose Multiple Semantic-Guided Context Optimization (MSGCoOp), a framework to enhance few-shot generalization while maintaining computational efficiency. MSGCoOp employs an ensemble of parallel learnable context vectors to capture diverse class semantics and introduces a semantic guidance mechanism using comprehensive class descriptions automatically generated by a Large Language Model (LLM) to enrich prompts. Additionally, a diversity regularization loss encourages prompts to learn complementary and orthogonal features, preventing redundant representations. Extensive experiments demonstrate that MSGCoOp improves base-to-novel generalization by $\text{1. 1 0 \%}$ on 11 datasets and achieves a 0.30 % performance gain in cross-domain tasks over the KgCoOp baseline. Our code is available at: https://github.com/Rain-Bus/MSGCoOp.