ViM: Out-Of-Distribution with Virtual-logit Matching

By Haoqi Wang, Zhizhong Li, Litong Feng, Wayne Zhang. 2022

549 citations recorded by the source corpus.

Most of the existing Out-Of-Distribution (OOD) detection algorithms depend on single input source: the feature, the logit, or the softmax probability. However, the immense diversity of the OOD examples makes such methods fragile. There are OOD samples that are easy to identify in the feature space while hard to distinguish in the logit space and vice versa. Motivated by this observation, we propose a novel OOD scoring method named Virtual-logit Matching (ViM), which combines the class-agnostic score from feature space and the In-Distribution (ID) class-dependent logits. Specifically, an additional logit representing the virtual OOD class is generated from the residual of the feature against the principal space, and then matched with the original logits by a constant scaling. The probability of this virtual logit after softmax is the indicator of OOD-ness. To facilitate the evaluation of large-scale OOD detection in academia, we create a new OOD dataset for ImageNet1K, which is human-annotated and is 8.8× the size of existing datasets. We conducted extensive experiments, including CNNs and vision transformers, to demonstrate the effectiveness of the proposed ViM score. In particular, using the BiT-S model, our method gets an average AUROC 90.91% on four difficult OOD benchmarks, which is 4% ahead of the best baseline. Code and dataset are available at https://github.com/haoqiwang/vim.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

Citations in the corpus

Learnability and Algorithm for Continual Learning (2023, background): “As pre-training models or feature extractors are increasingly used in all types of applications, including continual learning (Wang et al., 2022c; Kim et al., 2022a; Wu et al., 2022), we also take this approach.”
OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, methodology): “To build OOD dataset, we use a 10k subset of Species [51] with 713k images, iNaturalist [52] with 10k images, ImageNet-O [53] with 2k images, and OpenImage-O [54] with 17k images.”
Generalized Out-of-Distribution Detection: A Survey (2021, unknown): “In general, inference-only methods outperform trained methods.”
Out-of-distribution Detection with Deep Nearest Neighbors (2022, background): “…et al., 2017), Mahalanobis distance-based score (Lee et al., 2018), energy score (Liu et al., 2020; Lin et al., 2021; Wang et al., 2021; Morteza & Li, 2022), activation rectiﬁcation (ReAct) (Sun et al., 2021), gradient-based score (Huang et al., 2021) and ViM score (Wang et al., 2022).”