MaPLe: Multi-modal Prompt Learning

TLDR

Pre‑trained vision‑language models such as CLIP generalize well but are highly sensitive to prompt choice, and adapting only one modality limits the ability to jointly tune vision and language representations. This work proposes Multi‑modal Prompt Learning (MaPLe) to jointly learn prompts for both vision and language branches, enhancing their alignment. MaPLe couples vision and language prompts, discourages independent solutions, and learns stage‑wise prompts to progressively capture feature relationships, evaluated on generalization, dataset shift, and domain shift tasks. MaPLe outperforms Co‑CoOp, achieving a 3.45 % absolute gain on novel classes and 2.72 % on overall harmonic mean across 11 datasets, with code and pretrained models released online.

Abstract

Pre-trained vision-language (V-L) models such as CLIP have shown excellent generalization ability to downstream tasks. However, they are sensitive to the choice of input text prompts and require careful selection of prompt templates to perform well. Inspired by the Natural Language Processing (NLP) literature, recent CLIP adaptation approaches learn prompts as the textual inputs to fine-tune CLIP for downstream tasks. We note that using prompting to adapt representations in a single branch of CLIP (language or vision) is sub-optimal since it does not allow the flexibility to dynamically adjust both representation spaces on a downstream task. In this work, we propose Multi-modal Prompt Learning (MaPLe) for both vision and language branches to improve alignment between the vision and language representations. Our design promotes strong coupling between the vision-language prompts to ensure mutual synergy and discourages learning independent uni-modal solutions. Further, we learn separate prompts across different early stages to progressively model the stage-wise feature relationships to allow rich context learning. We evaluate the effectiveness of our approach on three representative tasks of generalization to novel classes, new target datasets and unseen domain shifts. Compared with the state-of-the-art method Co-CoOp, MaPLe exhibits favorable performance and achieves an absolute gain of 3.45% on novel classes and 2.72% on overall harmonic-mean, averaged over 11 diverse image recognition datasets. Our code and pre-trained models are available at https://github.com/muzairkhattak/multimodal-prompt-learning.