# Learning Open Set Network with Discriminative Reciprocal Points Source: [Learning Open Set Network with Discriminative Reciprocal Points]() Collection index: [llms.txt]() Data updated: 2025-10-25T22:49:52.437Z Authors: Guangyao Chen, Limeng Qiao, Yemin Shi, Peixi Peng, Jia Li, Tiejun Huang, Shiliang Pu, Yonghong Tian Published: 2020-10-31T03:20:31.000Z ## Abstract Open set recognition is an emerging research area that aims to simultaneously classify samples from predefined classes and identify the rest as 'unknown'. In this process, one of the key challenges is to reduce the risk of generalizing the inherent characteristics of numerous unknown samples learned from a small amount of known data. In this paper, we propose a new concept, Reciprocal Point, which is the potential representation of the extra-class space corresponding to each known category. The sample can be classified to known or unknown by the otherness with reciprocal points. To tackle the open set problem, we offer a novel open space risk regularization term. Based on the bounded space constructed by reciprocal points, the risk of unknown is reduced through multi-category interaction. The novel learning framework called Reciprocal Point Learning (RPL), which can indirectly introduce the unknown information into the learner with only known classes, so as to learn more compact and discriminative representations. Moreover, we further construct a new large-scale challenging aircraft dataset for open set recognition: Aircraft 300 (Air-300). Extensive experiments on multiple benchmark datasets indicate that our framework is significantly superior to other existing approaches and achieves state-of-the-art performance on standard open set benchmarks. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Open Set Recognition on CIFAR+50](): Open Set Recognition; CIFAR+50; AUROC: 96.8. - [Open-set recognition on CIFAR+50 (4 closed, 50 open classes)](): Open Set Recognition; CIFAR+50 1.0 (4 closed CIFAR10 classes, 50 open CIFAR100 classes); AUROC: 96.8. - [Open Set Recognition on CIFAR10](): Open Set Recognition; CIFAR10; AUROC: 0.976. - [Open-set recognition on CIFAR+10 (4 closed, 10 open from CIFAR100)](): Open Set Recognition; CIFAR+10 4 closed CIFAR10 classes, 10 open CIFAR100 classes 1.0; AUROC: 97.6. - [Open-Set Recognition on BreaKHis t=5 (test)](): Open Set Recognition; BreaKHis t=5 (test); Accuracy: 84.04. - [Open-Set Recognition on BloodMNIST t=5 (test)](): Open Set Recognition; BloodMNIST t=5 (test); Accuracy: 98. - [Open-Set Recognition on OCTMNIST t=3 (test)](): Open Set Recognition; OCTMNIST t=3 (test); Accuracy: 93.7. - [Open Set Action Recognition on UCF-101 + MiT-v2](): Open set action recognition; UCF-101 + MiT-v2; Open Set AUC: 77.42. - [Open-set object recognition on CIFAR+50 (4 closed, 50 open classes)](): Open-set object recognition; CIFAR+50 4 closed-set (from CIFAR10) and 50 open-set classes (from CIFAR100); OSCR: 89.6. - [Open Set Action Recognition on UCF-101 + HMDB-51](): Open set action recognition; UCF-101 + HMDB-51; Open Set AUC: 74.23. - [Open-Set Recognition on DermaMNIST v2 (test)](): Open Set Recognition; DermaMNIST v=2 (test); Accuracy: 80.785. - [Open-set object recognition on TinyImageNet (20 closed, 180 open classes)](): Open-set object recognition; TinyImageNet 20 closed-set and 180 open-set classes; OSCR: 53.2. - [Open Set Recognition on SVHN](): Open Set Recognition; SVHN; AUROC: 0.968. - [Open Set Recognition on TinyImageNet](): Open Set Recognition; TinyImageNet; AUROC: 80.9. - [Open-set recognition on TinyImageNet (20 closed, 180 open classes)](): Open Set Recognition; TinyImageNet 20 closed, 180 open classes 1.0; AUROC: 80.9. - [Open-set Recognition on CIFAR10 (6 closed, 4 open classes)](): Open Set Recognition; CIFAR10 6 closed, 4 open classes 1.0; AUROC: 0.901. - [Action Recognition on UCF101](): Action Recognition; UCF101; Accuracy: not recorded. ## Original sources and implementation [Original paper]() [PDF]()