Research map / Label-efficient and robust learning / Recognition and long-tail
Model Calibration in Dense Classification with Adaptive Label Perturbation
Jiawei Liu, Changkun Ye, Shan Wang, Ruikai Cui, Jing Zhang, Kaihao Zhang, Nick Barnes
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 changed | Where | Measured | Effect |
|---|---|---|---|
| make label perturbation probability adaptive per training image (ASLP_MC) instead of static SLP (ablation) | Stochastic Label Perturbation · DUTS-TE | ECE (%): 2.21 -> 1.40 | Improved |
| add stochastic label perturbation to baseline (ablation) | baseline SOD model · DUTS-TE | ECE (%): 3.48 -> 2.21 | Improved |
| use calibration regulariser (ASLP_MC) instead of accuracy regulariser (ASLP_MEI) (ablation) | ASLP · DUTS-TE | ECE (%): 27.9 -> 1.40 | Improved |
| use accuracy regulariser (ASLP_MEI) instead of calibration regulariser (ablation) | ASLP · Describable Textures (OOD) | ECE (%): 18.31 -> 13.43 | Improved |
| apply ASLP_MC instead of AdaFocal | AdaFocal · Describable Textures (OOD) | ECE (%): 27.55 -> 18.31 | Improved |
| apply temperature scaling | baseline SOD model · Describable Textures (OOD) | ECE (%): 52.36 -> 51.95 | No clear effect |
| make label smoothing factor adaptive per image (ALS) instead of static LS (ablation) | Label Smoothing · DUTS-TE | ECE (%): 2.20 -> 1.46 | Improved |
| add Gaussian noise (dynamic moderation) to moderation perturbation (ablation) | ASLP_MC · DUTS-TE | ECE (%): 1.47 -> 1.64 | Hurt |
| add dynamic Gaussian noise to moderation perturbation (ablation) | ASLP_MC · SOD dataset | ECE (%): 5.63 -> 4.14 | Improved |
| increase static label perturbation probability beyond a small value (ablation) | SLP · SOD testing datasets | ECE | Hurt |
| set regularisation strength lambda very high (ablation) | ASLP_MC · SOD | calibration | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdaFocal — AdaFocal: Calibration-aware Adaptive Focal Loss (NeurIPS 2022) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Calibrated Multi-label Deep Neural Networks (CVPR 2024 · similarity 0.86)
- The Minimax Rate of Second-Order Calibration (NeurIPS 2026 · similarity 0.85)
- Dirichlet-Based Prediction Calibration for Learning with Noisy Labels (AAAI 2024 · similarity 0.84)
- Beyond One-Hot Labels: Semantic Mixing for Model Calibration (ICML 2025 · similarity 0.84)
- Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · similarity 0.83)
- On Temperature Scaling and Conformal Prediction of Deep Classifiers (ICML 2025 · similarity 0.83)
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.
