Layer-Specific Prompt Fusion Discovery via Differentiable Search in Vision Foundation Models

2026-06-24

AI Summary

This work addresses the limitation of existing vision prompt tuning methods, which rely on a single image-prompt fusion strategy. The authors formulate the selection of fusion mechanisms as a bilevel optimization problem and employ differentiable architecture search to jointly optimize prompts and their layer-specific fusion schemes across Transformer layers. They introduce novel fusion operations combining affine transformations and cross-attention, enabling the first automatic discovery of heterogeneous, layer-adaptive fusion strategies. Extensive experiments across 34 datasets demonstrate that the proposed method significantly outperforms current prompt tuning baselines when using a frozen ViT backbone, achieving a superior trade-off among accuracy, inference latency, and parameter efficiency.

Abstract

Visual prompt tuning has emerged as a parameter-efficient fine-tuning approach for adapting large-scale Vision Transformers (ViTs) to downstream tasks. As its learnable prompts are applied in input and feature spaces, prior to jointly going through attention in transformer layers, the most commonly used scheme for fusing image and prompt tokens is concatenation or addition. In this paper, we aim to study a fundamental yet essential problem in visual prompt tuning: whether a single fusion scheme tends to yield better results, and whether that would be beneficial to develop a hybrid fusion scheme. To this end, we formulate the task as a bi-level optimization problem, and solve it leveraging differentiable architecture search. In this context, the learnable prompts and their fusion schemes are jointly optimized. To enrich the search space in the architecture search, we propose two additional fusion schemes, namely, affine transformation and cross-attention, in addition to concatenation and addition. Extensive experiments on 34 datasets spanning VTAB-1k, FGVC, and HTA show consistent gains over prompt-tuning baselines. With a frozen ViT backbone, our method delivers a favorable accuracy--latency--parameter trade-off compared with VPT-Deep and recent variants. Our findings reveal that how prompts fuse with image tokens plays a significant role in visual prompt tuning, and a hybrid fusion fashion can more effectively leverage layer semantics of ViTs, contributing a novel perspective for visual prompt-tuning research.

Problem

• visual prompt tuning
• prompt fusion
• Vision Transformers
• fusion scheme
• layer semantics

Innovation

• prompt fusion
• differentiable architecture search
• vision foundation models
• layer-specific optimization
• visual prompt tuning