arXiv 2603.151662026-03-16cs.CV

DAIT: 从视觉-语言模型到轻量分类器的适应性中间教师知识蒸馏

DAIT: Distillation from Vision-Language Models to Lightweight Classifiers with Adaptive Intermediate Teacher Transfer

School of Computer and Electronic Information, Nanjing Normal University, Nanjing, China（南京师范大学计算机与电子信息学院）；Medical Artificial Intelligence Lab, Westlake University（西湖大学医学人工智能实验室）

Zhengxu He, Jun Li, Zhijian Wu

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出DAIT方法，通过引入可训练的中间教师模型，从冻结的视觉-语言模型中适应性地蒸馏出任务对齐的知识，提升细粒度视觉分类性能。

摘要：

大规模视觉-语言模型（VLMs）编码了丰富的多模态语义，对细粒度视觉分类（FGVC）具有高度的益处。然而，其巨大的计算成本阻碍了在资源受限环境中的实际部署。尽管知识蒸馏有助于将VLMs的能力转移到轻量级分类器，但传统蒸馏机制直接从通用VLM转移到紧凑的学生模型，往往由于严重的架构错位和引入无关任务信息而产生次优结果。为缓解这一限制，本文提出Distillation with Adaptive Intermediate Teacher transfer（DAIT），促进从VLMs到轻量学生模型的适应性知识转移。DAIT引入了一个可训练的中间教师，该教师在显式监督下学习从目标细粒度任务转移冻结的VLM表示。该中间教师适应性地增强辨别性视觉线索，从而产生紧凑且任务对齐的知识，可以可靠地蒸馏到轻量模型中。在多个具有不同学生架构的FGVC基准测试中进行的广泛评估表明，本文方法在FGVC-Aircraft和CUB-200-2011数据集上分别实现了12.63%和8.34%的性能提升，确立了DAIT作为从通用VLMS到可部署细粒度识别模型的系统性范式。

英文摘要：

Large-scale Vision-Language Models (VLMs) encode rich multimodal semantics that are highly beneficial for fine-grained visual categorization (FGVC). However, their prohibitive computational cost hinders practical deployment in resource-constrained environments. Although knowledge distillation contributes to transferring VLMs capacity to lightweight classifiers, conventional distillation mechanisms, which directly transfer from a generic VLM to a compact student, often yield suboptimal results due to severe architectural misalignment and introducing task-irrelevant information. To alleviate this limitation, we propose Distillation with Adaptive Intermediate Teacher transfer (DAIT) in this study, facilitating adaptive knowledge transfer from VLMs to lightweight students. DAIT introduces a trainable intermediate teacher that learns to transfer frozen VLMs representations under explicit supervision from the target fine-grained task. This intermediate teacher adaptively enhances discriminative visual cues, thereby producing compact and task-aligned knowledge that can be reliably distilled into lightweight models. Extensive evaluations on multiple FGVC benchmarks with diverse student architectures demonstrate that our method achieves respective performance gains of 12.63% and 8.34% on FGVC-Aircraft and CUB-200-2011 datasets, establishing DAIT as a principled paradigm for transferring from general-purpose VLMS to deployable fine-grained recognition models.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。