Supervision Exists Everywhere: A Data Efficient Contrastive Language-Image Pre-training Paradigm

TL;DR

DeCLIP introduces a data-efficient contrastive learning paradigm that leverages multiple supervision types to improve visual feature learning, achieving comparable or better performance than CLIP with significantly less data.

Contribution

The paper proposes DeCLIP, a novel training paradigm that fully exploits supervision signals within and across modalities to reduce data requirements for effective contrastive pre-training.

Findings

DeCLIP-ResNet50 achieves 60.4% zero-shot ImageNet accuracy, surpassing CLIP-ResNet50.

DeCLIP requires 7.1 times less data than CLIP for similar performance.

Outperforms in 8 out of 11 downstream visual tasks.

Abstract

Recently, large-scale Contrastive Language-Image Pre-training (CLIP) has attracted unprecedented attention for its impressive zero-shot recognition ability and excellent transferability to downstream tasks. However, CLIP is quite data-hungry and requires 400M image-text pairs for pre-training, thereby restricting its adoption. This work proposes a novel training paradigm, Data efficient CLIP (DeCLIP), to alleviate this limitation. We demonstrate that by carefully utilizing the widespread supervision among the image-text pairs, our De-CLIP can learn generic visual features more efficiently. Instead of using the single image-text contrastive supervision, we fully exploit data potential through the use of (1) self-supervision within each modality; (2) multi-view supervision across modalities; (3) nearest-neighbor supervision from other similar pairs. Benefiting from intrinsic supervision, our DeCLIP-ResNet50 can achieve 60.4% zero-shot top1 accuracy on ImageNet, which is 0.8% above the CLIP-ResNet50 while using 7.1 x fewer data. Our DeCLIP-ResNet50 outperforms its counterpart in 8 out of 11 visual datasets when transferred to downstream tasks. Moreover, Scaling up the model and computing also works well in our framework.Our code, dataset and models are released at: https://github.com/Sense-GVT/DeCLIP