Open World Semi-supervised Learning Based on Multi-scale Enhanced Feature

Semi-supervised learning (SSL) is one of the main approaches to address the high cost of manual annotation in supervised learning. In recent years, SSL methods have effectively utilized consistency regularization on unlabeled data to improve performance while leveraging a small portion of labeled data as anchors. A common assumption in most SSL methods is that the unlabeled data generally share similar distributions and structures. However, this assumption does not align with many real-world scenarios. We lack knowledge about the underlying data distribution of unlabeled data. Most existing SSL methods are limited in their applicability to real-world problems. In contrast, in this work, we aim to address the challenging problem of open-world SSL, where the objective is to identify samples belonging to known classes while detecting and clustering samples from novel classes that appear in the unlabeled data. This paper introduces a method to discover new classes based on comparing multi-scale enhancement features. Extensive experiments demonstrate that our method outperforms the current state-of-the-art methods on multiple popular classification benchmarks while providing a better trade-off between accuracy and training time.