A Systematic Comparison of Training Objectives for Out-of-Distribution Detection in Image Classification
Not on the map. Why?
Abstract (arXiv)
Out-of-distribution (OOD) detection is critical in safety-sensitive applications. While this challenge has been addressed from various perspectives, the influence of training objectives on OOD behavior remains comparatively underexplored. In this paper, we present a systematic comparison of four widely used training objectives: Cross-Entropy Loss, Prototype Loss, Triplet Loss, and Average Precision (AP) Loss, spanning probabilistic, prototype-based, metric-learning, and ranking-based supervision, for OOD detection in image classification under standardized OpenOOD protocols. Within the evaluated ResNet-18/OpenOOD setting and objective-specific OOD scoring rules, Cross-Entropy Loss, Prototype Loss, and AP Loss achieve comparable in-distribution accuracy, while Cross-Entropy Loss provides the most consistent near- and far-OOD AUROC overall; the other objectives can be competitive in specific settings.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 32 of its 52 references that are in the map. First posted on arXiv 8 Mar 2026.
Builds on
32 of 52 references.
References · 32
- 2015Deep Residual Learning for Image Recognitioncited by 51,809
- 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
- 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
- 2017Prototypical Networks for Few-shot Learningcited by 3,001
- 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,833
- 2016A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networkscited by 2,362
- 2018A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attackscited by 1,472
- 2017Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networkscited by 1,325
- 2020Energy-based Out-of-distribution Detectioncited by 1,296
- 2018Deep Anomaly Detection with Outlier Exposurecited by 939
- 2019AugMix: A Simple Data Processing Method to Improve Robustness and Uncertaintycited by 897
- 2021Generalized Out-of-Distribution Detection: A Surveycited by 796
20 more
- 2017In Defense of the Triplet Loss for Person Re-Identificationcited by 784
- 2022Out-of-Distribution Detection with Deep Nearest Neighborscited by 596
- 2020CSI: Novelty Detection via Contrastive Learning on Distributionally Shifted Instancescited by 441
- 2022ViM: Out-Of-Distribution with Virtual-logit Matchingcited by 404
- 2020Calibrating Deep Neural Networks using Focal Losscited by 353
- 2020SCAN: Learning to Classify Images without Labelscited by 351
- 2021Open-Set Recognition: a Good Closed-Set Classifier is All You Need?cited by 351
- 2022OpenOOD: Benchmarking Generalized Out-of-Distribution Detectioncited by 267
- 2023OpenOOD v1.5: Enhanced Benchmark for Out-of-Distribution Detectioncited by 178
- 2023In or Out? Fixing ImageNet Out-of-Distribution Detection Evaluationcited by 132
- 2022How to Exploit Hyperspherical Embeddings for Out-of-Distribution Detection?cited by 128
- 2018Robust Classification with Convolutional Prototype Learningcited by 122
- 2020Smooth-AP: Smoothing the Path Towards Large-Scale Image Retrievalcited by 98
- 2019Few-Shot Learning with Embedded Class Models and Shot-Free Meta Trainingcited by 57
- 2023Rethinking Out-of-distribution (OOD) Detection: Masked Image Modeling is All You Needcited by 46
- 2024Learning with Mixture of Prototypes for Out-of-Distribution Detectioncited by 43
- 2019Towards Accurate One-Stage Object Detection with AP-Losscited by 36
- 2021Rank & Sort Loss for Object Detection and Instance Segmentationcited by 22
- 2024RankED: Addressing Imbalance and Uncertainty in Edge Detection Using Ranking-based Lossescited by 18
- 2024Geometric Insights into Focal Loss: Reducing Curvature for Enhanced Model Calibrationcited by 2
Built on by
Not cited yet.