Research map / Detection and segmentation / CNN features
Structure-Aware Representation Distillation for Tiny-Dense Object Segmentation
Xuesong Liu, Anke Xu, Wenbo Cao, Emmett Ientilucci
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 logit-level KD with uniform feature representation distillation (ablation) | ResNet-50 student, Swin-L teacher · RockFrag | mIoU: 52.3 -> 53.8 | Improved |
| add structure-aware weighting (E+C+D) to uniform SARD (ablation) | SARD ResNet-50 student · RockFrag | mIoU: 53.8 -> 58.6 | Improved |
| add structure-aware weighting (E+C+D) to uniform SARD (ablation) | SARD ResNet-50 student · RockFrag | bIoU: 38.5 -> 45.2 | Improved |
| use boundary-only weighting instead of full structure score (ablation) | SARD · RockFrag | mIoU: 58.6 -> 56.2 | Hurt |
| use density-only weighting instead of full structure score (ablation) | SARD · RockFrag | mIoU: 58.6 -> 55.4 | Hurt |
| add density term D to E+C weighting (ablation) | SARD · RockFrag | mIoU: 57.8 -> 58.6 | Improved |
| use SARD instead of CWD, Swin-L to Swin-T | Swin-T student · RockFrag | mIoU: 55.9 -> 60.2 | Improved |
| use SARD instead of CWD, Swin-L to Swin-T | Swin-T student · Cityscapes | mIoU: 78.9 -> 80.3 | Improved |
| use SARD instead of CWD, Swin-B to Swin-S | Swin-S student · Cityscapes | mIoU: 79.4 -> 80.1 | Improved |
| use SARD instead of CWD, Swin-B to Swin-S | Swin-S student · RockFrag | bIoU: 42.8 -> 46.5 | Improved |
| use SARD instead of CWD across heterogeneous teacher-student pairs | ViT-T, EfficientSAM-Ti, M2F-R50 students · Cityscapes/ADE20K/RockFrag average | mIoU: +1.5 to +1.8 mIoU, +2.3 to +2.7 bIoU | Improved |
| distill ResNet-50 with SARD instead of CWD | ResNet-50 student · RockFrag | mIoU: 54.1 -> 58.6 | Improved |
Closest papers by meaning
- Uncertainty-aware Fine-tuning of Segmentation Foundation Models (NeurIPS 2024 · similarity 0.85)
- $\mathbf{D^{3}S^{2}}$: Diffusion-Guided Dataset Distillation for Semantic Segmentation (NeurIPS 2026 · similarity 0.85)
- Near, far: Patch-ordering enhances vision foundation models' scene understanding (ICLR 2025 · similarity 0.85)
- High-Precision Dichotomous Image Segmentation via Probing Diffusion Capacity (ICLR 2025 · similarity 0.85)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.85)
- Multi-view Aggregation Network for Dichotomous Image Segmentation (CVPR 2024 · 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.
