Author: Zen and the Art of Computer Programming
1 Introduction
In the field of artificial intelligence, deep learning technology has gradually become an important means to solve complex tasks. However, there are still many difficulties in applying deep learning technology to the field of computer vision. Especially in practical application scenarios, the goal of computer vision is often to identify specific objects or scenes in images. Many existing deep learning models, such as CNN, VGG, etc., can handle general image classification tasks, but the performance of these models in real environments does not necessarily meet the requirements. For example, for those complex scenes, there may be some noise effects or even occlusion, which will cause the accuracy of the model to decrease. How to use weakly supervised data to improve model performance has also become a current research hotspot. This article explains the relevant principles, methods and applications of semi-supervised learning in computer vision, focusing on the following three aspects:
- 1. Overview of semi-supervised learning
- (1) What is semi-supervised learning?
- (2) What are the advantages and disadvantages of semi-supervised learning?
- (3) Types and representative models of semi-supervised learning.
- 2. Application of semi-supervised learning in computer vision
- (4) How to define weakly supervised data?
- (5) What is the specific operation process of semi-supervised learning models (such as SVM, Self-training, etc.)?
- (6) What are the specific principles, advantages and disadvantages of the semi-supervised learning framework used in this article - Causal-aware Unsupervised Domain Adaptation (CAUDA)?
- (7) What are the specific principles, advantages and disadvantages of the algorithm used in this article - Cross Entropy Method (CEM)?
- (8) Finally, this article shares the specific code implementation used in this article.