This repository is a curated collection of resources on Parameter-Efficient Transfer Learning (PEFT) for pre-trained vision models. It targets researchers and practitioners in computer vision who aim to achieve high performance with minimal parameter updates, offering a systematic review and categorization of existing PEFT methods.

How It Works

The project categorizes PEFT methods into three main groups: Addition-based Tuning, Partial-based Tuning, and Unified-based Tuning. This classification provides a structured overview of techniques that modify or add a small number of parameters to pre-trained models, enabling efficient adaptation to downstream tasks without the computational cost of full fine-tuning.

Quick Start & Requirements

This repository is a collection of papers and resources, not a runnable library. It requires no installation.

Highlighted Details

• Comprehensive categorization of PEFT methods including Adapter Tuning, Prompt Tuning, Prefix Tuning, Side Tuning, and Reparameter Tuning.
• Links to papers and code for numerous PEFT techniques.
• Includes a table of commonly used datasets for evaluating PEFT methods in various vision tasks.
• Features a survey paper "Parameter-Efficient Fine-Tuning for Pre-Trained Vision Models: A Survey" released in February 2024.

Maintenance & Community

The repository is marked as "Maintained? - yes". The latest commit was recent, indicating active curation.

Licensing & Compatibility

The repository itself does not have a specified license, but it links to numerous research papers and code repositories, each with its own licensing. Users must consult the licenses of individual linked projects.

Limitations & Caveats

This is a curated list of resources and does not provide a unified framework or benchmark for direct use. Users need to individually find, install, and integrate the PEFT methods from the linked code repositories.