Towards Optimal Feature-Shaping Methods for Out-of-Distribution Detection

Authors: Qinyu Zhao, Ming Xu, Kartik Gupta, Akshay Asthana, Liang Zheng, Stephen Gould

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments, we show that the feature-shaping function optimized by our method improves the generalization ability of OOD detection across a large variety of datasets and model architectures.
Researcher Affiliation Collaboration Qinyu Zhao1, Ming Xu1, Kartik Gupta2, Akshay Asthana2, Liang Zheng1, Stephen Gould1 1 The Australian National University 2 Seeing Machines Ltd
Pseudocode No The paper describes methods verbally and with equations, but does not include structured pseudocode or algorithm blocks with labels like 'Algorithm' or 'Pseudocode'.
Open Source Code Yes 1Our code is available at https://github.com/Qinyu-Allen-Zhao/OptFSOOD.
Open Datasets Yes We use CIFAR10, CIFAR100 (Krizhevsky et al., 2009), and Image Net-1k (Russakovsky et al., 2015) as ID datasets.
Dataset Splits Yes We extract ISFI vectors for the Image Net-1k training set (ID) (Russakovsky et al., 2015) and for the i Naturalist dataset (OOD) (Van Horn et al., 2018), using a Res Net50 model pre-trained on Image Net-1k training set. Specifically, we utilize Image Net-1k validation set (ID) and i Naturalist (OOD) with Res Net50.
Hardware Specification Yes All experiments are run on a single NVIDIA Ge Force RTX 4090 GPU.
Software Dependencies No The paper mentions software like 'Py Torch' and specific models but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Hyperparameters are set consistent with the original papers and concrete hyperparameter settings can be found in Section A.1 of the Appendix.