Classification-Reconstruction Learning for Open-Set Recognition

By Ryota Yoshihashi, Wen Shao, Rei Kawakami, Shaodi You, Makoto Iida, Takeshi Naemura. 2018

393 citations recorded by the source corpus.

Open-set classification is a problem of handling ‘unknown’ classes that are not contained in the training dataset, whereas traditional classifiers assume that only known classes appear in the test environment. Existing open-set classifiers rely on deep networks trained in a supervised manner on known classes in the training set; this causes specialization of learned representations to known classes and makes it hard to distinguish unknowns from knowns. In contrast, we train networks for joint classification and reconstruction of input data. This enhances the learned representation so as to preserve information useful for separating unknowns from knowns, as well as to discriminate classes of knowns. Our novel Classification-Reconstruction learning for Open-Set Recognition (CROSR) utilizes latent representations for reconstruction and enables robust unknown detection without harming the known-class classification accuracy. Extensive experiments reveal that the proposed method outperforms existing deep open-set classifiers in multiple standard datasets and is robust to diverse outliers.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Han-Jia Ye
2. De-chuan Zhan
3. Da-Wei Zhou
4. Ziwei Liu
5. Qiwen Wang
6. Zhiyuan Qi
7. Zhi-Hong Qi
8. C. Fendley

Citations in the corpus

• Few-Shot Continual Active Learning by a Robot (2022, methodology): “Another related field to active learning is open-set recognition (or out of the distribution detection (OOD)) [37, 38, 38, 39], in which a model that is trained on a set of tasks/classes might face samples belonging previously unknown classes in the test set.”
• A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, methodology): “More recent works have come to the conclusion that although the latter approaches have a strong theoretical foundation for open set recognition, they are still limited by activation values in discriminative neural networks being optimized exclusively towards predicting a correct class (Yoshihashi et”
• Recent Advances in Open Set Recognition: A Survey (2018, methodology): “Table 4 summaries the comparisons among these methods, where SoftMax [89], OpenMax [79], CROSR [85], and C2AE [87] from Deep Neural Network-based category, while G-OpenMax [88] and OSRCI [89] from Instance Generation-based category.”
• CODEs: Chamfer Out-of-Distribution Examples against Overconfidence Issue (2021, background): “All above approaches detect whether a test sample is from in-distribution (i.e., training distribution by a classifier) or OOD, and is usually deployed with combining the original n -category classifier to handle recognition in the real world [4, 46].”
• Generative-Discriminative Feature Representations for Open-Set Recognition (2020, methodology): “The performance of the baseline methods is obtained from [29] and [16].”
• Learning Bounds for Open-Set Learning (2021, methodology): “Following Yoshihashi et al. (2019), the macro-average F1 scores are used to evaluate OSL.”
• SSD: A Unified Framework for Self-Supervised Outlier Detection (2021, methodology): “Supervised detectors have been most successful with complex input modalities, such as images and language [4, 9, 10, 22, 48, 28].”
• Conditional Gaussian Distribution Learning for Open Set Recognition (2020, methodology): “Secondly, following the protocol defined in [37], all samples in CIFAR-10 dataset are collected as known data, and samples from other datasets, i.”
• End-to-End Reconstruction-Classification Learning for Face Forgery Detection (2022, background): “The objective of it is to model the distribution of input data in the embedding space [28, 34, 54].”
• Deep Class-Incremental Learning: A Survey (2023, background): “However, the performance of generative replay methods relies on the quality of generated data, which may fail in complex, large-scale inputs [125].”
• Hybrid Models for Open Set Recognition (2020, methodology): “The proposed OpenHybrid model outperforms both state-of-the-art methods [2,7,22,24,37] and hybrid model baselines [20,10] in these benchmarks.”
• Better the Devil you Know: An Analysis of Evasion Attacks using Out-of-Distribution Adversarial Examples (2019, background): “Indeed, there have been many recent attempts extending the state-of-the-art deep learning systems to the open-world environment [9, 10, 49, 131].”
• Generalized Out-of-Distribution Detection: A Survey (2021, neutral): “CROSR (Yoshihashi et al., 2019) enhances the features by concatenating visual embeddings from both the classifier and reconstruction model for distance computation in the extended feature space.”
• Unsupervised Out-of-Distribution Detection by Maximum Classifier Discrepancy (2019, result): “There are some other studies on open-set classification [1, 2, 6, 21, 23, 24, 30], which involve tasks very similar to OOD detection.”
• Class-Incremental Learning: A Survey (2023, background): “They are found to work well on simple datasets [166], [167] while failing in complex, large-scale inputs [168].”