Research map / Detection and segmentation / CNN features
Snapshot Distillation: Teacher-Student Optimization in One Generation
Chenglin Yang, Lingxi Xie, Chi Su, Alan L. Yuille
Check 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 |
|---|---|---|---|
| Use snapshot distillation instead of direct baseline training | DenseNet190 · CIFAR100 test | classification error (%): 16.69 -> 16.06 | Improved |
| Use snapshot distillation with T=3 instead of T=2 (ablation) | DenseNet190 · CIFAR100 test | classification error (%): 16.06 -> 16.33 | Hurt |
| Use snapshot distillation instead of direct baseline training | ResNet20 · CIFAR100 test | classification error (%): 33.57 -> 31.76 | Improved |
| Use snapshot distillation instead of snapshot ensemble | DenseNet190 · CIFAR100 test | classification error (%): 16.70 -> 16.06 | Improved |
| Use snapshot distillation instead of direct baseline training | ResNet101 · ILSVRC2012 test | Top-1 error (%): 21.62 -> 21.25 | Improved |
| Use snapshot distillation instead of direct baseline training | ResNet101 · ILSVRC2012 test | Top-5 error (%): 5.80 -> 5.55 | Improved |
| Use snapshot distillation instead of direct baseline training | ResNet152 · ILSVRC2012 test | Top-1 error (%): 21.17 -> 20.93 | Improved |
| Use snapshot distillation instead of direct baseline training | ResNet152 · ILSVRC2012 test | Top-5 error (%): 5.66 -> 5.55 | Improved |
| Use snapshot distillation instead of snapshot ensemble | ResNet152 · ILSVRC2012 test | Top-1 error (%): 21.84 -> 20.93 | Improved |
| Use snapshot distillation instead of snapshot ensemble | ResNet101 · ILSVRC2012 test | Top-1 error (%): 22.14 -> 21.25 | Improved |
| Use a pretrained backbone optimized with snapshot distillation rather than the baseline | ResNet152 · PascalVOC 2007 | mAP (%): 73.49 -> 74.93 | Improved |
| Use a pretrained backbone optimized with snapshot distillation rather than the baseline | ResNet152 · PascalVOC 2012 | mIOU (%): 77.53 -> 77.97 | Improved |
Closest papers by meaning
- Knowledge Distillation and Student-Teacher Learning for Visual Intelligence: A Review and New Outlooks (TPAMI 2021 · similarity 0.89)
- Learning Lightweight Object Detectors via Multi-Teacher Progressive Distillation (ICML 2023 · similarity 0.89)
- IDeaL: Data-Free Multi-Teacher Distillation via Improved Dead Leaves (ECCV 2026 · similarity 0.89)
- Long Short-Term Sample Distillation (AAAI 2020 · similarity 0.89)
- Training Deep Neural Networks in Generations: A More Tolerant Teacher Educates Better Students (AAAI 2019 · similarity 0.88)
- SDPGO: Efficient Self-Distillation Training Meets Proximal Gradient Optimization (NeurIPS 2025 · 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.
