Research map / Detection and segmentation / CNN features
Multi-Scale Aligned Distillation for Low-Resolution Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply the proposed aligned multi-scale fusion and distillation framework | FCOS, ResNet-50 · COCO object detection, L | AP: 38.2 -> 41.6 | Helped |
| Apply the proposed aligned multi-scale fusion and distillation framework | RetinaNet, ResNet-50 · COCO object detection, L | AP: 37.2 -> 40.3 | Helped |
| Apply the proposed aligned multi-scale fusion and distillation framework | Mask R-CNN, ResNet-50 · COCO instance segmentation, L | AP: 35.2 -> 37.3 | Helped |
| Apply the proposed aligned multi-scale fusion and distillation framework | Mask R-CNN, ResNet-50 · COCO keypoint detection, L | AP: 62.9 -> 65.0 | Helped |
| Add feature-map alignment to multi-scale training (ablation) | FCOS teacher · COCO val, low-resolution input | AP: 35.9 -> 37.8 | Helped |
| Replace vanilla multi-scale training with aligned multi-scale training (ablation) | FCOS teacher · COCO val, high-resolution input | AP: 40.3 -> 40.1 | No clear effect |
| Add crossing feature-level fusion instead of using a single-resolution feature set (ablation) | FCOS fusion teacher · COCO val | AP: 40.1 -> 42.5 | Helped |
| Use dynamic crossing fusion rather than static concatenation-and-convolution fusion (ablation) | FCOS fusion teacher · COCO val | AP: 41.4 -> 42.5 | Helped |
| Replace softmax fusion weights with sigmoid weights (ablation) | C-FF fusion teacher · COCO val | AP: 42.5 -> 41.6 | Hurt |
| Generate channel-wise fusion weights (ablation) | C-FF fusion teacher · COCO val | AP: 42.5 -> 42.5 | No clear effect |
| Use a multi-resolution C-FF teacher rather than a single-scale high-resolution teacher (ablation) | Low-resolution FCOS student · COCO val | AP: 38.2 -> 39.7 | Helped |
| Align features for output-level knowledge distillation (ablation) | Low-resolution FCOS student · COCO val | AP: 36.4 -> 37.5 | Helped |
Closest papers by meaning
- DFD: Distilling the Feature Disparity Differently for Detectors (ICML 2024 · similarity 0.90)
- Learning Lightweight Object Detectors via Multi-Teacher Progressive Distillation (ICML 2023 · similarity 0.90)
- Improve Object Detection with Feature-based Knowledge Distillation: Towards Accurate and Efficient Detectors (ICLR 2021 · similarity 0.89)
- Focal and Global Knowledge Distillation for Detectors (CVPR 2022 · similarity 0.89)
- DCSF-KD: Dynamic Channel-wise Spatial Feature Knowledge Distillation for Object Detection (AAAI 2025 · similarity 0.87)
- ScaleKD: Strong Vision Transformers Could Be Excellent Teachers (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.