Cluster-level pseudo-labelling forsource-free cross-domain facial expressionrecognition

Cluster-level pseudo-labelling for source-free cross-domain facial expression recognitionhttps://arxiv.org/abs/2210.05246文章的链接BVMC 2022

有代码，代码链接为：GitHub - altndrr/clup: Code implementation of our BMVC 2022 paper: Cluster-level pseudo-labelling for source-free cross-domain facial expression recognition

 首先根据source model 对target 数据进预测，设置shredhold, 如没有达到分界线，则根据self-pretrain 进行重新分配。

常见的source-free 算法：

1、SHOT [23] employs an entropy loss alongside a classification loss on pseudo-labelled samples to adapt the network to the target domain. The work has been extended in [ 24] introducing an auxiliary head that solves relative rotation, leading to improved performance. Differently from the above, the authors of [13] frame the problem from an image translation perspective and translate the target images to the source style using only the source model. In [36], they perform self-training with a loss function that considers the intrinsic structure of the target domain via nearest neighbours.

SHOT 在伪标记样本的分类损失的同时使用了熵损失，使网络适应目标域。
SHOT++扩展了SHOT, 解决相对旋转问题的辅助头，可提高性能。
不同的基于上述观点，[13]的作者从图像翻译的角度对问题进行了分析, 并且仅使用源模型将目标图像转换为源样式。
在[36]中，他们使用考虑目标内在结构的损失函数进行自我训练，target domain 的label 使用了最近邻居的方法进行确定。