This paper presents a comparative analysis of the application of Variational Autoencoders (VAE) and Wasserstein Generative Adversarial Networks (WGAN) for detecting anomalous images. The performance evaluation of the models is conducted using metrics such as AUC, Precision, Recall, and F1-Score. The results indicate that VAEs provide high Recall in anomaly detection but with low Precision, whereas WGANs demonstrate more balanced results with fewer false positives. Recommendations are proposed for selecting an appropriate model depending on the specific task requirements and the need to minimize false alarms or achieve high detection completeness.
Comparative Analysis of the Use of Variational Autoencoders and Generative Adversarial Networks for Anomalous Image Detection Tasks
2024-10-22
3399705 byte
Conference paper
Electronic Resource
English
MIDI-based generative neural networks with variational autoencoders for innovative music creation
BASE | 2024
|