Research map / Detection and segmentation / CNN features
Structured Knowledge Distillation for Semantic Segmentation
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 |
|---|---|---|---|
| Add pixel-wise distillation (ablation) | ResNet18 (0.5) · Cityscapes validation | mIoU (%): 55.37 ± 0.25 -> 57.07 ± 0.69 | Helped |
| Add pair-wise distillation after pixel-wise distillation (ablation) | ResNet18 (0.5) · Cityscapes validation | mIoU (%): 57.07 ± 0.69 -> 61.03 ± 0.49 | Helped |
| Add holistic distillation after pixel-wise and pair-wise distillation (ablation) | ResNet18 (0.5) · Cityscapes validation | mIoU (%): 61.03 ± 0.49 -> 61.63 ± 0.99 | Helped |
| Add ImageNet-pretrained initialization (ablation) | ResNet18 (1.0) with pixel-wise, pair-wise, and holistic distillation · Cityscapes validation | mIoU (%): 63.24 ± 0.74 -> 72.67 ± 0.57 | Helped |
| Use global pair-wise distillation instead of MIMIC feature distillation (ablation) | ResNet18 (1.0), ImageNet-pretrained, with pixel-wise distillation · Cityscapes validation | mIoU (%): 71.03 -> 71.37 | Helped |
| Use global pair-wise distillation instead of attention transfer (ablation) | ResNet18 (1.0), ImageNet-pretrained, with pixel-wise distillation · Cityscapes validation | mIoU (%): 70.70 -> 71.37 | Helped |
| Use global pair-wise distillation instead of local pair-wise distillation (ablation) | ResNet18 (1.0), ImageNet-pretrained, with pixel-wise distillation · Cityscapes validation | mIoU (%): 70.86 -> 71.37 | Helped |
| Add distillation | ESPNet · CamVid test | mIoU (%): 57.8 -> 61.4 | Helped |
| Add extra unlabeled Cityscapes street-scene data for distillation | ESPNet · CamVid test | mIoU (%): 61.4 -> 65.1 | Helped |
| Add extra unlabeled Cityscapes street-scene data for distillation | ESPNet-C · CamVid test | mIoU (%): 60.3 -> 64.1 | Helped |
| Add pixel-wise, pair-wise, and holistic distillation | ResNet18 · ADE20K validation | mIoU (%): 33.82 -> 36.55 | Helped |
| Add pixel-wise, pair-wise, and holistic distillation | MobileNetV2Plus · ADE20K validation | mIoU (%): 33.64 -> 35.51 | Helped |
Closest papers by meaning
- Knowledge Adaptation for Efficient Semantic Segmentation (CVPR 2019 · similarity 0.90)
- Multi-Knowledge Aggregation and Transfer for Semantic Segmentation (AAAI 2022 · similarity 0.90)
- Distilling Knowledge from Heterogeneous Architectures for Semantic Segmentation (AAAI 2025 · similarity 0.90)
- Augmentation-Free Dense Contrastive Knowledge Distillation for Efficient Semantic Segmentation (NeurIPS 2023 · similarity 0.89)
- Generalizable Knowledge Distillation from Vision Foundation Models for Semantic Segmentation (CVPR 2026 · similarity 0.89)
- Knowledge Distillation from A Stronger Teacher (NeurIPS 2022 · 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.