Generative OpenMax for Multi-Class Open Set Classification

By ZongYuan Ge, Sergey Demyanov, Zetao Chen, Rahil Garnavi. 2017

469 citations recorded by the source corpus.

We present a conceptually new and flexible method for multi-class open set classification. Unlike previous methods where unknown classes are inferred with respect to the feature or decision distance to the known classes, our approach is able to provide explicit modelling and decision score for unknown classes. The proposed method, called Gener- ative OpenMax (G-OpenMax), extends OpenMax by employing generative adversarial networks (GANs) for novel category image synthesis. We validate the proposed method on two datasets of handwritten digits and characters, resulting in superior results over previous deep learning based method OpenMax Moreover, G-OpenMax provides a way to visualize samples representing the unknown classes from open space. Our simple and effective approach could serve as a new direction to tackle the challenging multi-class open set classification problem.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Media Technology

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

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

Towards Open World Object Detection (2021, background): “A generative version of OpenMax was proposed in [13] by synthesizing novel class images.”
Learning Placeholders for Open-Set Recognition (2021, methodology): “We compare to the SOTA methods, i.e ., Softmax, OpenMax [3], G-OpenMax [6], OSRCI [22], C2AE [24], CROSR [36], GFROSR [26]: Softmax : utilizes the highest softmax probability as the conﬁdence score for detection; OpenMax : replaces softmax layer with OpenMax and cal-ibrates the conﬁdence to predict…”
Recent Advances in Open Set Recognition: A Survey (2018, methodology): “Due to the properties of AL, some researchers also attempt to account for open space with the unknown unknown classes generated by the AL technique [78]–[81].”
C2AE: Class Conditioned Auto-Encoder for Open-Set Recognition (2019, methodology): “However, these questions are relatively unexplored in the context of deep neural networks [38], [4], [24], [8], [7].”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “[160] [180] [181] [153] [154] [155] [173] [183] [184]”
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.”
OpenGAN: Open-Set Recognition via Open Data Generation (2021, unknown): “These additional images are used to augment the training set for learning an open-set model, which oftentimes is designed for both the closed-world K-way classification and open-set discrimination [18, 27, 33].”
CMG: A Class-Mixed Generation Approach to Out-of-Distribution Detection (2022, methodology): “In some of these methods, generated data are treated as OOD samples to optimize the decision boundary and calibrate the confidence [60, 54].”
OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, methodology): “In fact, methods developed for OSR [27, 28], model uncertainty [29, 30], and even data augmentation methods [31] can seamlessly solve the OOD detection problem.”
Classification-Reconstruction Learning for Open-Set Recognition (2018, methodology): “G-Openmax [10], a direct extension of Openmax, trains networks with synthesized unknown data by using generative models.”
Open Set Domain Adaptation by Backpropagation (2018, methodology): “Moreover, to give supervision of the unknown samples, a method to generate these samples was proposed [26].”
Generative-Discriminative Feature Representations for Open-Set Recognition (2020, methodology): “In [7], a Generative Adversarial Network (GAN) based framework was used to estimate open-set class activations.”
WildFish: A Large Benchmark for Fish Recognition in the Wild (2018, background): “…Non-Outlier (NNO) [3], etc. Recently, the remarkable successes of convolutional neural networks (CNNs) [9, 18, 20, 24, 37, 38, 45] have promoted open-set deep learning [4, 6, 16, 36] and its relevant applications such as person re-identification [51], face recognition [17], scene parsing [49].”
Learning Bounds for Open-Set Learning (2021, methodology): “Further, to adapt deep networks support to OSL, Bendale & Boult (2016), Ge et al. (2017) proposed OpenMax and G-OpenMax, respectively.”
Hybrid Models for Open Set Recognition (2020, methodology): “Since then, the OpenMax is further developed in [25,6].”
Few-Shot Open-Set Recognition Using Meta-Learning (2020, methodology): “We compare the proposed method to several openset recognition SOTA methods, including OpenMax [1], G-OpenMax [5], and Counterfactual [21], and an out-ofdistribution SOTA method, Confidence [14].”
Open Set Learning with Counterfactual Images (2018, methodology): “We interpolate between the two examples in the latent space as in [7] and apply the generator to the resulting latent point to generate the image:”
Generalized Category Discovery (2022, background): “GANs are often employed to generate adversarial samples to train an open-set classiﬁer, e.g, [14, 25, 32].”