The paper introduces ODIN (Out-of-Distribution detector for Neural networks), a simple and effective method to detect out-of-distribution images in neural networks without requiring any retraining of the pre-trained model. ODIN combines temperature scaling and input preprocessing to enhance the softmax score distribution between in-distribution and out-of-distribution images. The method is shown to be compatible with various network architectures and datasets, outperforming the baseline approach (Hendrycks & Gimpel, 2017) by a significant margin. Experiments on state-of-the-art architectures like DenseNet and Wide ResNet demonstrate that ODIN reduces the false positive rate from 34.7% to 4.3% on the DenseNet (applied to CIFAR-10 and Tiny-ImageNet) when the true positive rate is 95%. The paper also provides an empirical analysis of parameter settings and discusses the intuition behind the method, including the effects of temperature scaling and input preprocessing.The paper introduces ODIN (Out-of-Distribution detector for Neural networks), a simple and effective method to detect out-of-distribution images in neural networks without requiring any retraining of the pre-trained model. ODIN combines temperature scaling and input preprocessing to enhance the softmax score distribution between in-distribution and out-of-distribution images. The method is shown to be compatible with various network architectures and datasets, outperforming the baseline approach (Hendrycks & Gimpel, 2017) by a significant margin. Experiments on state-of-the-art architectures like DenseNet and Wide ResNet demonstrate that ODIN reduces the false positive rate from 34.7% to 4.3% on the DenseNet (applied to CIFAR-10 and Tiny-ImageNet) when the true positive rate is 95%. The paper also provides an empirical analysis of parameter settings and discusses the intuition behind the method, including the effects of temperature scaling and input preprocessing.