Raising the bar on the evaluation of out-of-distribution detection

Conference item

In image classification, a lot of development has happened in detecting out-of-distribution (OoD) data. However, most OoD detection methods are evaluated on a standard set of datasets, arbitrarily different from training data. There is no clear definition of what forms a "good" OoD dataset. Furthermore, the state-of-the-art OoD detection methods already achieve near perfect results on these standard benchmarks. In this paper, we define 2 categories of OoD data using the subtly different concepts of perceptual/visual and semantic similarity to in-distribution (iD) data. We define Near OoD samples as perceptually similar but semantically different from iD samples, and Shifted samples as points which are visually different but semantically akin to iD data. We then propose a GAN based framework for generating OoD samples from each of these 2 categories, given an iD dataset. Through extensive experiments on MNIST, CIFAR-10/100 and ImageNet, we show that a) state-of-the-art OoD detection methods which perform exceedingly well on conventional benchmarks are significantly less robust to our proposed benchmark. Moreover, we observe that b) models performing well on our setup also perform well on conventional real-world OoD detection benchmarks and vice versa, thereby indicating that one might not even need a separate OoD set, to reliably evaluate performance in OoD detection.

Authors: Mukhoti, J; Lin, T-Y; Chen, B-C; Shah, A; Torr, PHS

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: EEE
• Host title: 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)
• Pages: 4367-4377
• Publication date: 2023-12-25
• Acceptance date: 2023-10-06
• Event title: 2023 International Conference on Computer Vision Workshops (ICCV Workshops)
• Event location: Paris
• Event website: https://iccv2023.thecvf.com/
• Event start date: 2023-10-02
• Event end date: 2023-10-06
• DOI: 10.1109/iccvw60793.2023.00471
• EISSN: 2473-9944
• ISSN: 2473-9936
• EISBN: 9798350307443
• ISBN: 9798350307450
• Language: English
• Keywords: training data; benchmark testing; computational modeling; semantics; computer vision; natural languages; conferences