CLIP Models are Few-shot Learners: Empirical Studies on VQA and Visual Entailment

TL;DR

This paper demonstrates that CLIP, originally a visual encoder, possesses strong zero-shot and few-shot capabilities for vision-language tasks like VQA and visual entailment, without extra pre-training.

Contribution

It empirically shows CLIP's few-shot abilities for vision-language tasks and proposes a parameter-efficient fine-tuning method to enhance performance.

Findings

CLIP achieves competitive zero-shot VQA results.

CLIP demonstrates zero-shot transfer to visual entailment.

Fine-tuning improves few-shot VQA performance.

Abstract

CLIP has shown a remarkable zero-shot capability on a wide range of vision tasks. Previously, CLIP is only regarded as a powerful visual encoder. However, after being pre-trained by language supervision from a large amount of image-caption pairs, CLIP itself should also have acquired some few-shot abilities for vision-language tasks. In this work, we empirically show that CLIP can be a strong vision-language few-shot learner by leveraging the power of language. We first evaluate CLIP's zero-shot performance on a typical visual question answering task and demonstrate a zero-shot cross-modality transfer capability of CLIP on the visual entailment task. Then we propose a parameter-efficient fine-tuning strategy to boost the few-shot performance on the vqa task. We achieve competitive zero/few-shot results on the visual question answering and visual entailment tasks without introducing any additional pre-training procedure.