Leveraging Vision-Language Models as Weak Annotators in Active Learning

TL;DR

This paper explores using vision-language models as weak annotators to reduce human labeling effort in active learning, especially by leveraging their strength in coarse-grained labeling.

Contribution

It introduces a novel active learning framework combining human fine-grained labels with VLM-generated coarse labels, modeling label noise with trusted data.

Findings

The framework outperforms existing active learning methods on CUB200 and FGVC-Aircraft datasets.

VLMs perform well on coarse labels but poorly on fine-grained labels.

Combining human and VLM labels reduces annotation costs effectively.

Abstract

Active learning aims to reduce annotation cost by selectively querying informative samples for supervision under a limited labeling budget. In this work, we investigate how vision-language models (VLMs) can be leveraged to further reduce the reliance on costly human annotation within the active learning paradigm. To this end, we find that the reliability of VLMs varies significantly with label granularity in fine-grained recognition tasks: they perform poorly on fine-grained labels but can provide accurate coarse-grained labels. Leveraging this property, we propose an active learning framework that combines fine-grained human annotations with coarse-grained VLM-generated weak labels through instance-wise label assignment. We further model the systematic noise in VLM-generated labels using a small set of trusted full labels. Experiments on CUB200 and FGVC-Aircraft show that the proposed framework consistently outperforms existing active learning methods under the same annotation budget.