Research map / Label-efficient and robust learning / Recognition and long-tail
Open Set Learning with Counterfactual Images
Lawrence Neal, Matthew Olson, Xiaoli Fern, Weng-Keen Wong, Fuxin Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add counterfactual open set images as an extra class | softmax classifier · SVHN open set detection (K=6, M=10) | ROC AUC: .886 -> .910 | Improved |
| add counterfactual open set images as an extra class | softmax classifier · CIFAR-10 open set detection (K=6, M=10) | ROC AUC: .677 -> .699 | Improved |
| add counterfactual open set images as an extra class | softmax classifier · MNIST open set detection (K=6, M=10) | ROC AUC: .978 -> .988 | Improved |
| use counterfactual generation instead of OpenMax | classifier · SVHN open set detection (K=6, M=10) | ROC AUC: .894 -> .910 | Improved |
| use counterfactual optimization instead of latent interpolation (G-OpenMax*) | synthetic open set generation · CIFAR-10 open set detection (K=6, M=10) | ROC AUC: .675 -> .699 | Improved |
| use counterfactual optimization instead of latent interpolation (G-OpenMax*) | synthetic open set generation · Tiny-ImageNet extended open set | ROC AUC: .580 -> .586 | Improved |
| use counterfactual optimization instead of latent interpolation (G-OpenMax*) | synthetic open set generation · CIFAR+50 extended open set | ROC AUC: .819 -> .827 | Improved |
| add counterfactual images to training, effect on known-class accuracy | classifier · CIFAR-10 closed set accuracy (K=6) | accuracy: .801 -> .821 | Improved |
| add counterfactual images to training, effect on known-class accuracy | classifier · MNIST closed set accuracy (K=6) | accuracy: .995 -> .996 | No clear effect |
| use generative OpenMax (interpolation) rather than OpenMax | OpenMax · CIFAR-10 open set detection (K=6, M=10) | ROC AUC: .695 -> .675 | No clear effect |
| use G-OpenMax* instead of OpenMax on digit datasets | OpenMax · MNIST open set detection (K=6, M=10) | ROC AUC: .981 -> .984 | Improved |
| use G-OpenMax* instead of OpenMax | OpenMax · CIFAR+50 extended open set | ROC AUC: .796 -> .819 | Improved |
Closest papers by meaning
- Generative-Discriminative Feature Representations for Open-Set Recognition (CVPR 2020 · similarity 0.91)
- Counterfactual Zero-Shot and Open-Set Visual Recognition (CVPR 2021 · similarity 0.86)
- OpenGAN: Open-Set Recognition via Open Data Generation (ICCV 2021 · similarity 0.86)
- Unlocking Better Closed-Set Alignment Based on Neural Collapse for Open-Set Recognition (AAAI 2025 · similarity 0.84)
- From Coarse to Fine-Grained Open-Set Recognition (CVPR 2024 · similarity 0.84)
- Frequency Shuffling and Enhancement for Open Set Recognition (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
