Deep Open Set Recognition Using Dynamic Intra-class Splitting

Abstract

This paper provides a generic deep learning method to solve open set recognition problems. In open set recognition, only samples of a limited number of known classes are given for training. During inference, an open set recognizer must not only correctly classify samples from known classes, but also reject samples from unknown classes. Due to these specific requirements, conventional deep learning models that assume a closed set environment cannot be used. Therefore, special open set approaches were taken, including variants of support vector machines and generation-based state-of-the-art methods which model unknown classes by generated samples. In contrast, our proposed method models unknown classes by atypical subsets of training samples. The subsets are obtained through intra-class splitting (ICS). Based on a recently proposed two-stage algorithm using ICS, we propose a one-stage method based on alternating between ICS and the training of a deep neural network. Finally, several experiments were conducted to compare our proposed method with conventional and other state-of-the-art methods. The proposed method based on dynamic ICS showed a comparable or better performance than all considered existing methods regarding balanced accuracy.

Towards open domain-specific recognition using Quad-Channel Self-Attention Reciprocal Point Learning and Autoencoder

“…This mechanism encourages the feature samples to reside between the true labels and the non-targeted labels. Schlachter et al [34] consider training images as open-set data, proposing a dynamic intraclass splitting technique that identifies atypical samples that are misclassified or yield low confidence scores.…”

Deep Metric Learning-Based Strawberry Disease Detection With Unknowns

“…Significant progress has been made with machine intelligence, which is another technique for continual and life-long learning for open-world recognition, even if it is premature for practical applications, especially fine-grained tasks ( Schlachter et al, 2019a , b , 2020 ; Geng et al, 2020 ). In the most general problem settings of the open world, no type of unknown can be contained in the training dataset, that is, it only appears in the test environment.…”

Quad-Channel Contrastive Prototype Networks for Open-Set Recognition in Domain-Specific Tasks

“…Unfortunately, most evaluations in OSR research have used experimental datasets, with MNIST, CIFAR10, CIFAR50, and/or TinyImageNet being the most popular [9], [10], [11], [12], [13], [14], [15], [16]. Only a few studies have used other datasets when constructing an open-set model [17], [18], [19], but these approaches may be ineffective when dealing with real-world applications in a specific domain.…”

“…In a study by Schlachter et al [13], the unknown set could be derived using a misclassified set or low probability prediction. This set provides additional unknown classes.…”