Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks

By Shiyu Liang, Yixuan Li, Rayadurgam Srikant. 2017

2,544 citations recorded by the source corpus.

We consider the problem of detecting out-of-distribution images in neural networks. We propose ODIN, a simple and effective method that does not require any change to a pre-trained neural network. Our method is based on the observation that using temperature scaling and adding small perturbations to the input can separate the softmax score distributions between in- and out-of-distribution images, allowing for more effective detection. We show in a series of experiments that ODIN is compatible with diverse network architectures and datasets. It consistently outperforms the baseline approach by a large margin, establishing a new state-of-the-art performance on this task. For example, ODIN reduces the false positive rate from the baseline 34.7% to 4.3% on the DenseNet (applied to CIFAR-10) when the true positive rate is 95%.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

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

In Safety, Robustness & Interpretability

Citations in the corpus

CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features (2019, methodology): “CutMix also enhances the model robustness and alleviates the over-confidence issue [13, 21] of deep networks.”
VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models (2020, methodology): “Following Lin et al. (2018), we report the number of covered modes and the KL divergence from the categorical distribution over 1000 categories from generated samples to true data (Table 5).”
Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift (2018, background): “[28] Shiyu Liang, Yixuan Li, and R Srikant.”
Deep Weighted Averaging Classifiers (2018, background): “There have also been several papers focused on the problem of predicting whether data is in-domain or out-of-domain [25, 28].”
A Theoretical Study on Solving Continual Learning (2022, methodology): “ODIN: Researchers have proposed several methods to improve the OOD detection performance of a trained network by post-processing [7, 70, 71].”
Detection and Continual Learning of Novel Face Presentation Attacks (2021, methodology): “Our idea is based on enabling the network to identify new attack types as testing samples that are outside the training distribution samples (OTDS) in an embedding space [11, 29, 40, 19, 4].”
Self-Supervised Training Enhances Online Continual Learning (2021, background): “We showed that self-supervised features generalize to both unseen classes and unseen datasets in online settings, so extending our work to open world learning would only require the addition of a component to identify if samples are unknown or out-of-distribution [41, 31, 36].”
Towards Open World Object Detection (2021, background): “In similar spirit, several dedicated approaches target on detecting the out of distribution samples [30] or novelties [48].”
Continual learning with hypernetworks (2019, background): “For categorical target distributions, the network ideally produces a flat, high entropy output for unseen data and, conversely, a peaked, low-entropy response for in-distribution data (Hendrycks & Gimpel, 2016; Liang et al., 2017).”
Being Bayesian, Even Just a Bit, Fixes Overconfidence in ReLU Networks (2020, background): “Post-hoc heuristics based on temperature or Platt scaling (Platt et al., 1999; Guo et al., 2017; Liang et al., 2018) are unable to detect inputs with arbitrarily high conﬁdence far away from the training data (Hein et al., 2019).”
Out-of-Domain Detection for Natural Language Understanding in Dialog Systems (2019, methodology): “imum value of the Softmax output as the detection score [11], which has been demonstrated to work surprisingly well on the image classification tasks [12], and thus has been applied in many state-of-the-art systems [12], [13].”
Why ReLU Networks Yield High-Confidence Predictions Far Away From the Training Data and How to Mitigate the Problem (2018, methodology): “In contrast to [22, 20] we neither use a diﬀerent parameter set for each test dataset [22] nor do we use one of the test datasets during training [20].”
Predictive Uncertainty Estimation via Prior Networks (2018, methodology): “The ﬁrst experiment illustrates the advantages of a DPN over other non-Bayesian methods [22, 30] on synthetic data and the second set of experiments evaluate DPNs on MNIST and CIFAR-10 and compares them to DNNs and ensembles generated via Monte-Carlo Dropout (MCDP) on the tasks of misclassiﬁcation…”
Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples (2017, methodology): “…the performance of the baseline detector for out-of-distribution using large-scale CNNs, i.e., VGGNets with 13 layers, under various training scenarios, where more results on AlexNet and ODIN detector (Liang et al., 2017) can be found in Appendix B (the overall trends of re-sults are similar).”