Research map / Label-efficient and robust learning / Knowledge distillation
Distilling Global and Local Logits With Densely Connected Relations
Youmin Kim, Jinbae Park, YounHo Jang, Muhammad Ali, Tae-Hyun Oh, Sung-Ho Bae
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 |
|---|---|---|---|
| replace KD with GLD (global + local logits + relations) | KD student WRN16-2 · CIFAR-100 val (WRN22-4 -> WRN16-2) | top-1 accuracy: 74.75 -> 76.28 | Improved |
| replace KD with GLD | ResNet18 student · ImageNet val | top-1 accuracy: 70.66 -> 71.63 | Improved |
| distill with GLD instead of no distillation, then transfer to CINIC-10 | WRN16-1 student · CIFAR-100 to CINIC-10 transfer | top-1 accuracy: 58.36 -> 61.09 | Improved |
| distill with GLD for SSD detector backbone | SSD-ResNet18 · COCO2017 val detection | mAP: 19.41 -> 19.83 | Improved |
| distill with GLD for DeepLabV3+ backbone | DLV3+-ResNet18 · COCO2017 val segmentation | mIoU: 51.22 -> 53.75 | Improved |
| use standard deviation instead of temperature 4 as softening factor, global logit only (ablation) | KD (global logit only) · CIFAR-100 val (case c) | top-1 accuracy: 74.75 -> 74.30 | Hurt |
| use standard deviation instead of temperature 4 as softening factor, global+local logits (ablation) | GLD without relation loss · CIFAR-100 val (cases a,b,c,d) | top-1 accuracy: 69.77 -> 71.02 (a), 67.44 -> 68.85 (b), 75.06 -> 74.89 (c), 72.50 -> 74.25 (d) | Mixed |
| increase local logits from 2x2 to 7x7 on fine-grained CUB-200-2011 (ablation) | GLD · CUB-200-2011 | top-1 accuracy: 68.89 -> 70.81 | Improved |
| increase local logits from 2x2 to 7x7 on ImageNet (ablation) | GLD · ImageNet val | top-1 accuracy: 71.63 -> 69.95 | Hurt |
| increase local logits from 2x2 to 8x8 on CIFAR-100 (case b) (ablation) | GLD · CIFAR-100 val (WRN16-2 -> WRN16-1) | top-1 accuracy: 68.95 -> 65.66 | Hurt |
| vary alpha in 0.1-0.7 and beta in 100-1000 (ablation) | GLD · CIFAR-100 val (case c) | top-1 accuracy: 76.00 to 76.28 | No clear effect |
Closest papers by meaning
- Decoupled Knowledge Distillation (CVPR 2022 · similarity 0.91)
- Structured Knowledge Distillation for Accurate and Efficient Object Detection (TPAMI 2023 · similarity 0.89)
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · similarity 0.89)
- Scaled Decoupled Distillation (CVPR 2024 · similarity 0.88)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.87)
- Wasserstein Distance Rivals Kullback-Leibler Divergence for Knowledge Distillation (NeurIPS 2024 · similarity 0.87)
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.
