PyramidCLIP: Hierarchical Feature Alignment for Vision-language Model Pretraining

TL;DR

PyramidCLIP introduces a hierarchical feature alignment method for vision-language pretraining that improves semantic matching and data efficiency, outperforming existing models on multiple downstream tasks.

Contribution

The paper proposes PyramidCLIP, a hierarchical alignment framework that addresses semantic mismatches and negative sample constraints in vision-language pretraining.

Findings

PyramidCLIP outperforms CLIP on ImageNet zero-shot classification by over 10% with the same data.

It achieves state-of-the-art results on several downstream tasks with larger datasets.

PyramidCLIP improves data efficiency, surpassing CLIP with fewer training pairs.

Abstract

Large-scale vision-language pre-training has achieved promising results on downstream tasks. Existing methods highly rely on the assumption that the image-text pairs crawled from the Internet are in perfect one-to-one correspondence. However, in real scenarios, this assumption can be difficult to hold: the text description, obtained by crawling the affiliated metadata of the image, often suffers from the semantic mismatch and the mutual compatibility. To address these issues, we introduce PyramidCLIP, which constructs an input pyramid with different semantic levels for each modality, and aligns visual elements and linguistic elements in the form of hierarchy via peer-level semantics alignment and cross-level relation alignment. Furthermore, we soften the loss of negative samples (unpaired samples) so as to weaken the strict constraint during the pre-training stage, thus mitigating the risk of forcing the model to distinguish compatible negative pairs. Experiments on five downstream tasks demonstrate the effectiveness of the proposed PyramidCLIP. In particular, with the same amount of 15 million pre-training image-text pairs, PyramidCLIP exceeds CLIP on ImageNet zero-shot classification top-1 accuracy by 10.6%/13.2%/10.0% with ResNet50/ViT-B32/ViT-B16 based image encoder respectively. When scaling to larger datasets, PyramidCLIP achieves the state-of-the-art results on several downstream tasks. In particular, the results of PyramidCLIP-ResNet50 trained on 143M image-text pairs surpass that of CLIP using 400M data on ImageNet zero-shot classification task, significantly improving the data efficiency of CLIP.