TiP-Adapter: Training-free CLIP-Adapter for Better Vision-Language Modeling

This is the official code release for the paper 'TiP-Adapter: Training-free CLIP-Adapter for Better Vision-Language Modeling'.

Introduction

Tip-Adapter provides faster convergence and better performance than CLIP-Adapter by initializing the adapter with a cache model.

Implementation

Put tip_adapter_ImageNet.py into clip's folder and run

python tip_adapter_ImageNet.py

you will get 65.51% on ImageNet validation set.

This repo will be completed in a few days.

Contributors

Peng Gao, Renrui Zhang

Acknowledgment

CLIP, CoOp and CLIP-Adapter