Research map / Label-efficient and robust learning / Recognition and long-tail
VICAL: Vicinal Consistency Alignment for Long-Tailed Visual Recognition
Jianggang Zhu, Zheng Wang, Bin Zhu, Yi-Ping Phoebe Chen, Jingjing Chen
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 |
|---|---|---|---|
| add self-consistency learning (SC) to multi-expert model (ablation) | multi-expert baseline · CIFAR-100-LT (IF=100) | top-1 accuracy: 55.6 -> 57.8 | Improved |
| add deep ensemble distillation (DED) (ablation) | ME + SC · CIFAR-100-LT (IF=100) | top-1 accuracy: 57.8 -> 58.6 | Improved |
| add conflicting knowledge filter (CKF) (ablation) | ME + SC + DED · CIFAR-100-LT (IF=100) | top-1 accuracy: 58.6 -> 59.1 | Improved |
| use low-resolution view for DED instead of full-resolution (ablation) | DED · CIFAR-100-LT (IF=100) | top-1 accuracy: 58.1 -> 59.1 | Improved |
| add low-pass filter to full-resolution input for DED (ablation) | DED · CIFAR-100-LT (IF=100) | top-1 accuracy: 58.1 -> 58.6 | Improved |
| add high-pass filter to full-resolution input for DED (ablation) | DED · CIFAR-100-LT (IF=100) | top-1 accuracy: 58.1 -> 58.1 | No clear effect |
| distill only on samples where teacher is wrong and student correct (ablation) | DED without CKF · CIFAR-100-LT (IF=100) | top-1 accuracy: 57.7 -> 53.2 | Hurt |
| use CKF instead of distilling on all samples (ablation) | DED · CIFAR-100-LT (IF=100) | top-1 accuracy: 58.6 -> 59.1 | Improved |
| use SC instead of vanilla self-distillation for a single expert (ablation) | single expert · CIFAR-100-LT (200 epochs) | top-1 accuracy: 51.7 -> 53.6 | Improved |
| train longer (400 vs 155 epochs) (ablation) | VICAL · CIFAR-100-LT (IF=100) | top-1 accuracy: 57.6 -> 59.7 | Improved |
| increase diversity regularizer weight lambda from 0 to 2 (ablation) | MDCS-style multi-expert · CIFAR-100-LT (IF=100) | top-1 accuracy: 49.9 -> 49.9 | No clear effect |
| use target network for inference instead of online network (ablation) | VICAL · CIFAR-100-LT (IF=100) | top-1 accuracy | Improved |
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.
- MDCS — MDCS: More Diverse Experts with Consistency Self-distillation for Long-tailed Recognition (ICCV 2023) · better on 1 of 1 shared result
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
- Global and Local Mixture Consistency Cumulative Learning for Long-Tailed Visual Recognitions (CVPR 2023 · similarity 0.89)
- MDCS: More Diverse Experts with Consistency Self-distillation for Long-tailed Recognition (ICCV 2023 · similarity 0.88)
- Distribution Alignment: A Unified Framework for Long-Tail Visual Recognition (CVPR 2021 · similarity 0.88)
- Trust-calibrated Collaborative Learning for Long-Tailed Visual Recognition (CVPR 2026 · similarity 0.87)
- Deep Long-Tailed Learning: A Survey (TPAMI 2023 · similarity 0.87)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · similarity 0.85)
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.
