Research map / Label-efficient and robust learning / Knowledge distillation
ACAM-KD: Adaptive and Cooperative Attention Masking for Knowledge Distillation
Qizhen Lan, Qing Tian
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 channel masking to spatial masking (ablation) | ACAM-KD · COCO RetinaNet R101->R50 | mAP: 40.9 -> 41.2 | Improved |
| add spatial masking to channel masking (ablation) | ACAM-KD · COCO RetinaNet R101->R50 | mAP: 40.4 -> 41.2 | Improved |
| teacher generates query instead of student (ablation) | STCA-FF · COCO RetinaNet R101->R50 | mAP: 41.0 -> 41.2 | Improved |
| add masking to distillation (fixed teacher masking) (ablation) | feature KD · COCO RetinaNet R101->R50 | mAP: 37.4 -> 39.8 | Improved |
| make teacher-driven mask adaptive during distillation (ablation) | fixed teacher masking · COCO RetinaNet R101->R50 | mAP: 39.8 -> 39.9 | No clear effect |
| add student-teacher cross-attention feature fusion to adaptive masking (ablation) | adaptive teacher masking · COCO RetinaNet R101->R50 | mAP: 39.9 -> 41.2 | Improved |
| add student-teacher cross-attention fusion and adaptive masking (ablation) | fixed teacher masking · COCO RetinaNet R101->R50 | AP_s: 21.5 -> 24.6 | Improved |
| use stronger ResNeXt-101 teacher instead of ResNet-101 (ablation) | ACAM-KD (RetinaNet) · COCO RetinaNet | mAP: 41.2 -> 41.5 | Mixed |
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.
- FreeKD — FreeKD: Knowledge Distillation via Semantic Frequency Prompt (CVPR 2024) · better on 5 of 5 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
- GLAMD: Global and Local Attention Mask Distillation for Object Detectors (ECCV 2022 · similarity 0.91)
- Masked Autoencoders Are Stronger Knowledge Distillers (ICCV 2023 · similarity 0.91)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.90)
- Knowledge Distillation and Student-Teacher Learning for Visual Intelligence: A Review and New Outlooks (TPAMI 2021 · similarity 0.90)
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.90)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.88)
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.
