LSD-C: Linearly Separable Deep Clusters –Supplementary Material–
ICCVW(2021)
摘要
Self-supervised pretraining. We train the RotNet [3] (i.e. predicting the rotation applied to the image among four possibilities: 0◦, 90◦, 180◦, and 270◦) on all datasets with the same configuration. Following the authors’ released code, we train for 200 epochs using a step-wise learning rate starting at 0.1 which is then divided by 5 at epochs 60, 120, and 160. Main LSD-C models. After the self-supervised pretraining step, following [4] we freeze the first three macro-blocks of the ResNet-18 [5] as the RotNet training provides robust early filters. We then train the last macro-block and the linear classifier using our clustering method. For all the experiments, we use a batch size of 256. We summarize in table 1 all the hyperparameters for the different datasets and labeling methods.
更多查看译文
关键词
data augmentation,semi-supervised learning practice,unlabeled dataset,linearly separable deep clusters,LSD-C,document classification dataset Reuters 10K,feature representation,binary cross-entropy loss,pairwise connections,similarity metric,feature space
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络