Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cascading Unknown Detection with Known Classification for Open Set Recognition

Daniel Brignac, Abhijit Mahalanobis

TL;DR

Cas-DC tackles open set recognition by decoupling unknown detection from known-class classification through a two-stage cascade. It trains an embedding network with a triplet loss to separate knowns from unknowns using a representative set of known unknowns, and a distance-based prototype discriminator to label samples as known or unknown, followed by a separate classifier for fine-grained known-class predictions. This specialization yields improved AUROC and CCR across multiple open-set benchmarks, indicating robust discrimination in both the open space and among known classes. The approach demonstrates practical impact by enabling more reliable deployment of models in open-world settings, where unknown instances must be detected before constrained classification.

Abstract

Deep learners tend to perform well when trained under the closed set assumption but struggle when deployed under open set conditions. This motivates the field of Open Set Recognition in which we seek to give deep learners the ability to recognize whether a data sample belongs to the known classes trained on or comes from the surrounding infinite world. Existing open set recognition methods typically rely upon a single function for the dual task of distinguishing between knowns and unknowns as well as making known class distinction. This dual process leaves performance on the table as the function is not specialized for either task. In this work, we introduce Cascading Unknown Detection with Known Classification (Cas-DC), where we instead learn specialized functions in a cascading fashion for both known/unknown detection and fine class classification amongst the world of knowns. Our experiments and analysis demonstrate that Cas-DC handily outperforms modern methods in open set recognition when compared using AUROC scores and correct classification rate at various true positive rates.

Cascading Unknown Detection with Known Classification for Open Set Recognition

TL;DR

Cas-DC tackles open set recognition by decoupling unknown detection from known-class classification through a two-stage cascade. It trains an embedding network with a triplet loss to separate knowns from unknowns using a representative set of known unknowns, and a distance-based prototype discriminator to label samples as known or unknown, followed by a separate classifier for fine-grained known-class predictions. This specialization yields improved AUROC and CCR across multiple open-set benchmarks, indicating robust discrimination in both the open space and among known classes. The approach demonstrates practical impact by enabling more reliable deployment of models in open-world settings, where unknown instances must be detected before constrained classification.

Abstract

Deep learners tend to perform well when trained under the closed set assumption but struggle when deployed under open set conditions. This motivates the field of Open Set Recognition in which we seek to give deep learners the ability to recognize whether a data sample belongs to the known classes trained on or comes from the surrounding infinite world. Existing open set recognition methods typically rely upon a single function for the dual task of distinguishing between knowns and unknowns as well as making known class distinction. This dual process leaves performance on the table as the function is not specialized for either task. In this work, we introduce Cascading Unknown Detection with Known Classification (Cas-DC), where we instead learn specialized functions in a cascading fashion for both known/unknown detection and fine class classification amongst the world of knowns. Our experiments and analysis demonstrate that Cas-DC handily outperforms modern methods in open set recognition when compared using AUROC scores and correct classification rate at various true positive rates.
Paper Structure (12 sections, 6 equations, 4 figures, 4 tables)

This paper contains 12 sections, 6 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Methodology
  4. Cascading Unknown Detection with Known Classification
  5. Embedding Separation of Knowns and Unknowns
  6. Discrimination Between Knowns and Unknowns
  7. Management of Open Space Risk
  8. Distinction Amongst Knowns
  9. Experiments and Results
  10. Experimental Setup
  11. Results Comparison
  12. Conclusion

Figures (4)

  • Figure 1: In conventional learning, classes from the known space (training classes) and the infinite surrounding image space map to the same decision space. Open set recognition methods allow the decision space to map back to the infinite surrounding image space by means of an "unknown" label.
  • Figure 2: Cas-DC training procedure (a) and inference procedure (b).
  • Figure 3: CIFAR10 TSNE plot of the embedding space.
  • Figure 4: CIFAR+10 CCR for varying TPR.