Efficient and Long-Tailed Generalization for Pre-trained Vision-Language Model

TL;DR

This paper introduces Candle, a framework that enhances pre-trained vision-language models like CLIP for long-tailed and zero-shot scenarios by using prototype-based methods, loss adjustments, and virtual prototypes, achieving state-of-the-art results efficiently.

Contribution

Candle is a novel framework that improves long-tailed and zero-shot generalization of CLIP by using logit-adjusted loss, cross-modal attention, and virtual prototypes for new classes.

Findings

Achieves state-of-the-art performance on 11 datasets.

Reduces training time significantly.

Effectively handles long-tailed and zero-shot tasks.

Abstract

Pre-trained vision-language models like CLIP have shown powerful zero-shot inference ability via image-text matching and prove to be strong few-shot learners in various downstream tasks. However, in real-world scenarios, adapting CLIP to downstream tasks may encounter the following challenges: 1) data may exhibit long-tailed data distributions and might not have abundant samples for all the classes; 2) There might be emerging tasks with new classes that contain no samples at all. To overcome them, we propose a novel framework to achieve efficient and long-tailed generalization, which can be termed as Candle. During the training process, we propose compensating logit-adjusted loss to encourage large margins of prototypes and alleviate imbalance both within the base classes and between the base and new classes. For efficient adaptation, we treat the CLIP model as a black box and leverage the extracted features to obtain visual and textual prototypes for prediction. To make full use of multi-modal information, we also propose cross-modal attention to enrich the features from both modalities. For effective generalization, we introduce virtual prototypes for new classes to make up for their lack of training images. Candle achieves state-of-the-art performance over extensive experiments on 11 diverse datasets while substantially reducing the training time, demonstrating the superiority of our approach. The source code is available at https://github.com/shijxcs/Candle.