AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Can multi-label classification networks know what they don’t know?

Haoran Wang, Weitang Liu, Alex Bocchieri, Yixuan Li

NeurIPS 2021 · Label-efficient and robust learning › OOD and anomaly detection · approach: Spurious-aware OOD detection · 35 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Replace MaxLogit with JointEnergyDenseNet-121 multi-label classifier · MS-COCO → ImageNet OODFPR95: 43.53 -> 33.48Improved
Replace Mahalanobis with JointEnergyDenseNet-121 multi-label classifier · PASCAL-VOC → ImageNet OODFPR95: 41.74 -> 41.01Improved
Replace MaxLogit with JointEnergyDenseNet-121 multi-label classifier · NUS-WIDE → ImageNet OODFPR95: 56.46 -> 48.98Improved
Sum label-wise energies rather than taking their maximum (ablation)JointEnergy · MS-COCO, PASCAL-VOC, and NUS-WIDE → ImageNet OODAUROC: 89.11 -> 92.70; 89.22 -> 91.10; 83.58 -> 88.30Improved
Sum logits across labels instead of using label-wise energies (ablation)Multi-label OOD scoring · MS-COCO → ImageNet OODFPR95: 43.53 -> 95.46Hurt
Sum label-wise probabilities instead of energies (ablation)Multi-label OOD scoring · MS-COCO → ImageNet OODFPR95: 33.48 -> 45.04Hurt
Sum label-wise probabilities instead of energies (ablation)Multi-label OOD scoring · PASCAL-VOC → ImageNet OODFPR95: 41.01 -> 38.57Improved
Sum label-wise probabilities instead of energies (ablation)Multi-label OOD scoring · NUS-WIDE → ImageNet OODFPR95: 48.98 -> 50.84Hurt
Restrict the sum to the top-k labels (ablation)JointEnergy · MS-COCO → ImageNet OODFPR95: 33.48 -> 37.85Hurt
Restrict the sum to the top-k labels (ablation)JointEnergy · PASCAL-VOC → ImageNet OODFPR95: 41.01 -> 43.63Hurt
Restrict the sum to the top-k labels (ablation)JointEnergy · NUS-WIDE → ImageNet OODFPR95: 48.98 -> 53.00Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy