UDAPose：面向低光照人体姿态估计的无监督域自适应
UDAPose: Unsupervised Domain Adaptation for Low-Light Human Pose Estimation
针对低光照姿态估计，用基于直流高通滤波（DHF）与低光照特征注入模块（LCIM）的合成管线生成逼真低光训练图像，并在 Transformer 中用动态注意力控制（DCA）在图像线索不可靠时回退到姿态先验——这套「合成域+注意力门控」思路可直接迁移到其他低质成像下的关键点/稠密预测任务。
原文依据 a Dynamic Control of Attention (DCA) module that adaptively balances image cues with learned pose priors in the Transformer architecture
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论低能见度场景（如低光照条件）给人 体姿态估计带来重大挑战，其原因在于带标注的低光照数据集稀缺，且不良光照下视觉信息丢失。近期的域自适应技术试图通过将光照良好的图像增广以模拟低光照条件，从而利用光照良好条件下的标签。但手工设计的增广过度简化了噪声模式，而基于学习的方法往往无法保留低光照的高频特性，生成的图像不真实，导致姿态模型在真实低光照场景中泛化性差。此外，近期的姿态估计器通过图像到关键点的交叉注意力依赖图像线索，但这些线索在低光照条件下变得不可靠。为解决上述问题，我们提出面向姿态估计的无监督域自适应框架 UDAPose，该框架合成低光照图像并动态融合视觉线索与姿态先验以改进姿态估计。具体而言，我们的合成方法引入基于直流的高通滤波器（DHF）与低光照特性注入模块（LCIM），从输入低光照图像中注入高频细节，克服了现有方法中的刚性或细节丢失问题。此外，我们引入注意力动态控制（DCA）模块，在 Transformer 架构中自适应地平衡图像线索与学习到的姿态先验。实验表明，UDAPose 优于最优方法，在 ExLPose-test 困难集（LL-H）上取得 10.1（56.4%）的显著 AP 提升，在 EHPT-XC 的跨数据集验证中提升 7.4（31.4%）。
英文原文摘要 1627 字符
Low-visibility scenarios, such as low-light conditions, pose significant challenges to human pose estimation due to the scarcity of annotated low-light datasets and the loss of visual information under poor illumination. Recent domain adaptation techniques attempt to utilize well-lit labels by augmenting well-lit images to mimic low-light conditions. But handcrafted augmentations oversimplify noise patterns, while learning-based methods often fail to preserve high-frequency low-light characteristics, producing unrealistic images that lead pose models to generalize poorly to real low-light scenes. Moreover, recent pose estimators rely on image cues through image-to-keypoint cross-attention, but these cues become unreliable under low-light conditions. To address these issues, we propose Unsupervised Domain Adaptation for Pose Estimation (UDAPose), a novel framework that synthesizes low-light images and dynamically fuses visual cues with pose priors for improved pose estimation. Specifically, our synthesis method incorporates a Direct-Current-based High-Pass Filter (DHF) and a Low-light Characteristics Injection Module (LCIM) to inject high-frequency details from input low-light images, overcoming rigidity or the detail loss in existing approaches. Furthermore, we introduce a Dynamic Control of Attention (DCA) module that adaptively balances image cues with learned pose priors in the Transformer architecture. Experiments show that UDAPose outperforms state-of-the-art methods, with notable AP gains of 10.1 (56.4%) on the ExLPose-test hard set (LL-H) and 7.4 (31.4%) in cross-dataset validation on EHPT-XC.