Introduction

As we know, "Dataset is expensive", semi-supervised learning(SSL) has proven to be successful in overcoming labeling difficulties by leveraging unlabeled data. However, currently there is a important problem while using SSL, Class-Imbalanced. Real-world datasets face usually this problem, some SSL algorithms ignore it and setup a fix threshold for all the classes while creating the pseudo-label, which will decrease the accuracy of prediction.

Introduction

(Before we start this blog, I must clarify that it serves as an initial introduction to the "Mean Teacher" framework. I may enhance this article later if I come across intriguing insights or new information.)

Introduction

An imbalanced dataset refers to a situation where the distribution of classes or labels in the dataset is highly skewed, meaning that one class (the minority class) is significantly underrepresented compared to another class (the majority class).

Introduction

The concept of the diffusion probability model was initially proposed by Jascha Sohl-Dickstein et al. in 2015. However, due to limitations in hardware devices such as memory at that time, this model did not receive much attention. Thanks to the development of technology, especially the advancements in GPUs and memory devices in recent years, the diffusion model has started to gain recognition.

Introduction

This blog post is a summary of a dataset with information that may be useful in current projects. And the paper is:

CaDIS: Cataract Dataset for Image Segmentation

Introduction

The swin transformer is proposed because the previous transformer has the following two shortcomings in the field of visual processing: