Research map / Label-efficient and robust learning / Recognition and long-tail
Shortcut-Resistant CAM Distillation for Long-Tailed Recognition
Wenhai Wan, Teng Zhang, Shao-Yuan Li, Xinrui Wang, Qiang-Sheng Hua, Songcan 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 SRCD to BCL | BCL · CIFAR-100-LT IR=100 (from scratch) | top-1 accuracy (%): 51.76 -> 53.64 | Improved |
| add SRCD to BCL | BCL · CIFAR-10-LT IR=100 (from scratch) | top-1 accuracy (%): 83.61 -> 82.97 | Hurt |
| add SRCD to GCL | GCL · CIFAR-100-LT IR=10 (from scratch) | top-1 accuracy (%): 61.79 -> 60.23 | Hurt |
| add SRCD to CE+H2T | CE+H2T · CIFAR-100-LT IR=50 (from scratch) | top-1 accuracy (%): 47.58 -> 46.97 | Hurt |
| add SRCD to GLMC | GLMC · ImageNet-LT Tail (from scratch) | top-1 accuracy (%): 30.41 -> 34.22 | Improved |
| add SRCD to LIFT (CLIP fine-tuning) | LIFT · ImageNet-LT All | top-1 accuracy (%): 77.1 -> 78.4 | Improved |
| add SRCD to LIFT (CLIP fine-tuning) | LIFT · iNaturalist Tail | top-1 accuracy (%): 81.5 -> 80.8 | Hurt |
| use both coherence and concentration cues in teacher weighting instead of one (ablation) | CE+FCC+SRCD · ImageNet-LT All | top-1 accuracy (%): 54.02 -> 55.95 | Improved |
| use energy-based weighting instead of simple averaging of head CAMs (ablation) | SRCD teacher · CAM similarity to balanced reference | CAM cosine similarity | Improved |
| increase number of head proxies K beyond 3 (ablation) | SRCD · long-tailed benchmark | accuracy | No clear effect |
| increase distillation strength beta above 0.6 (ablation) | SRCD · CIFAR-100-LT | accuracy | Improved |
| replace head-proxy teacher with fixed center mask (ablation) | SRCD · long-tailed benchmark | accuracy | 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.
- Focal-SAM — Focal-SAM: Focal Sharpness-Aware Minimization for Long-Tailed Classification (ICML 2025) · 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
- Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation (CVPR 2025 · similarity 0.88)
- DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta Debugging (NeurIPS 2026 · similarity 0.87)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.87)
- On the Foundations of Shortcut Learning (ICLR 2024 · similarity 0.87)
- OFBD: Object-Focused Background Debiasing for Long-Tailed Learning (NeurIPS 2026 · similarity 0.86)
- DiffuLT: Diffusion for Long-tail Recognition Without External Knowledge (NeurIPS 2024 · similarity 0.86)
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.
