OCFormer: One-Class Transformer Network for Image Classification

TL;DR

OCFormer introduces a novel Vision Transformer-based framework for one-class classification, utilizing Gaussian noise as pseudo-negatives, achieving superior results over CNN-based methods on multiple datasets.

Contribution

This work is the first to apply Vision Transformers to one-class classification, introducing a Gaussian noise pseudo-negative approach and an optimized loss function.

Findings

Significant performance improvements over CNN-based methods

Effective latent space representation using Gaussian noise

Validated on multiple benchmark datasets

Abstract

We propose a novel deep learning framework based on Vision Transformers (ViT) for one-class classification. The core idea is to use zero-centered Gaussian noise as a pseudo-negative class for latent space representation and then train the network using the optimal loss function. In prior works, there have been tremendous efforts to learn a good representation using varieties of loss functions, which ensures both discriminative and compact properties. The proposed one-class Vision Transformer (OCFormer) is exhaustively experimented on CIFAR-10, CIFAR-100, Fashion-MNIST and CelebA eyeglasses datasets. Our method has shown significant improvements over competing CNN based one-class classifier approaches.