Research map / Detection and segmentation / CNN features
Multi-Knowledge Aggregation and Transfer for Semantic Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Aggregate multiple intermediate knowledge types with ITE and APB | DeepLabV3-MobileNet · Pascal VOC 2012 | mIoU (%): 68.87 -> 71.86 | Helped |
| Aggregate multiple intermediate knowledge types with ITE and APB | DeepLabV3-ResNet18 · Pascal VOC 2012 | mIoU (%): 69.38 -> 72.42 | Helped |
| Aggregate multiple intermediate knowledge types with ITE and APB | DeepLabV3-ResNet18 · Cityscapes | mIoU (%): 67.77 -> 71.34 | Helped |
| Aggregate multiple intermediate knowledge types with ITE and APB | DeepLabV3-MobileNet · Cityscapes | mIoU (%): 71.61 -> 73.98 | Helped |
| Use MKAT instead of standard training | DeepLabV3-MobileNet · CamVid val | mIoU (%): 63.74 -> 65.18 | Helped |
| Use proxy online learning rather than fixed, pre-trained auxiliary modules (ablation) | MKAT · Cityscapes | mIoU: 70.85 -> 71.34 | Helped |
| Use proxy online learning rather than fixed, pre-trained auxiliary modules (ablation) | MKAT · CamVid | mIoU: 61.90 -> 62.19 | Helped |
| Add mutual label calibration (ablation) | MKAT · Cityscapes | mIoU: 70.89 -> 71.34 | Helped |
| Add mutual label calibration (ablation) | MKAT · CamVid | mIoU: 61.82 -> 62.19 | Helped |
| Replace mutual label calibration with whole-image KL soft-label loss (ablation) | MKAT · Cityscapes | mIoU: 71.34 -> 68.92 | Hurt |
| Replace mutual label calibration with whole-image KL soft-label loss (ablation) | MKAT · CamVid | mIoU: 62.19 -> 60.12 | Hurt |
| Increase the number of aggregated knowledge types (ablation) | MKAT · Pascal VOC, Cityscapes, and CamVid | mIoU | Helped |
Closest papers by meaning
- Structured Knowledge Distillation for Semantic Segmentation (CVPR 2019 · similarity 0.90)
- Distilling Knowledge from Heterogeneous Architectures for Semantic Segmentation (AAAI 2025 · similarity 0.90)
- Refine Myself by Teaching Myself: Feature Refinement via Self-Knowledge Distillation (CVPR 2021 · similarity 0.88)
- Knowledge Adaptation for Efficient Semantic Segmentation (CVPR 2019 · similarity 0.88)
- Self-Decoupling and Ensemble Distillation for Efficient Segmentation (AAAI 2023 · similarity 0.88)
- Inter-Region Affinity Distillation for Road Marking Segmentation (CVPR 2020 · 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.