arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2402.12736cs.CVcs.AI
收藏

CST:面向参数与内存高效迁移学习的校准侧调优

CST: Calibration Side-Tuning for Parameter and Memory Efficient Transfer Learning

Feng Chen

AI总结:

本文提出 Calibration side tuning 轻量级微调策略,将 adapter tuning 与 side tuning 结合并引入最大转移校准,以少量额外参数适配 ResNet,在五个基准数据集上超越现有方法,实现复杂度与性能的更好平衡。

摘要:

在目标检测中实现普遍较高的精度相当具有挑战性,当前工业界的主流关注点在于检测特定类别的物体。然而,部署一个或多个目标检测网络需要一定量的 GPU 内存用于训练,以及用于推理的存储容量。这在资源受限条件下如何有效协调多个目标检测任务方面带来了挑战。本文引入了一种轻量级微调策略,称为 Calibration side tuning(校准侧调优),该策略融合了 adapter tuning 和 side tuning 的若干方面,以将 transformer 中成功使用的技术适配到 ResNet 上。Calibration side tuning 架构引入了最大转移校准(maximal transition calibration),利用少量额外参数来提升网络性能,同时保持平滑的训练过程。此外,本文对多种微调策略进行了分析,并在 ResNet 中实现了它们的应用,从而扩展了目标检测网络微调策略的研究。另外,本文使用五个基准数据集进行了大量实验。实验结果表明,该方法优于其他对比的最先进技术,并且在微调方案的复杂度与性能之间取得了更好的平衡。

英文摘要:

Achieving a universally high accuracy in object detection is quite challenging, and the mainstream focus in the industry currently lies on detecting specific classes of objects. However, deploying one or multiple object detection networks requires a certain amount of GPU memory for training and storage capacity for inference. This presents challenges in terms of how to effectively coordinate multiple object detection tasks under resource-constrained conditions. This paper introduces a lightweight fine-tuning strategy called Calibration side tuning, which integrates aspects of adapter tuning and side tuning to adapt the successful techniques employed in transformers for use with ResNet. The Calibration side tuning architecture that incorporates maximal transition calibration, utilizing a small number of additional parameters to enhance network performance while maintaining a smooth training process. Furthermore, this paper has conducted an analysis on multiple fine-tuning strategies and have implemented their application within ResNet, thereby expanding the research on fine-tuning strategies for object detection networks. Besides, this paper carried out extensive experiments using five benchmark datasets. The experimental results demonstrated that this method outperforms other compared state-of-the-art techniques, and a better balance between the complexity and performance of the finetune schemes is achieved.

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。

↑