CPT: Colorful Prompt Tuning for Pre-trained Vision-Language Models

TL;DR

CPT introduces a novel prompt tuning method for vision-language models that reformulates visual grounding as a fill-in-the-blank task, significantly enhancing few-shot and zero-shot performance with minimal labeled data.

Contribution

The paper proposes Cross-modal Prompt Tuning (CPT), a new paradigm that bridges the gap between pre-training and fine-tuning in VL-PTMs using color-based co-referential prompts.

Findings

Outperforms fine-tuning methods with large accuracy gains.

Enables strong few-shot and zero-shot visual grounding capabilities.

Reduces standard deviation in performance, indicating more stable results.

Abstract

Pre-Trained Vision-Language Models (VL-PTMs) have shown promising capabilities in grounding natural language in image data, facilitating a broad variety of cross-modal tasks. However, we note that there exists a significant gap between the objective forms of model pre-training and fine-tuning, resulting in a need for large amounts of labeled data to stimulate the visual grounding capability of VL-PTMs for downstream tasks. To address the challenge, we present Cross-modal Prompt Tuning (CPT, alternatively, Colorful Prompt Tuning), a novel paradigm for tuning VL-PTMs, which reformulates visual grounding into a fill-in-the-blank problem with color-based co-referential markers in image and text, maximally mitigating the gap. In this way, CPT enables strong few-shot and even zero-shot visual grounding capabilities of VL-PTMs. Comprehensive experimental results show that the prompt-tuned VL-PTMs outperform their fine-tuned counterparts by a large margin (e.g., 17.3% absolute accuracy improvement, and 73.8% relative standard deviation reduction on average with one shot in RefCOCO evaluation). We make the data and code for this paper publicly available at https://github.com/thunlp/CPT.