Research map / Label-efficient and robust learning / Knowledge distillation
Knowledge Distillation for Learned Image Compression
Yunuo Chen, Zezheng Lyu, Bing He, Ning Cao, Gang Chen, Guo Lu, Wenjun Zhang
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 |
|---|---|---|---|
| add knowledge distillation to pruned student (ablation) | KDIC · Kodak | BD-rate (%): -11.21 -> -13.71 | Improved |
| add knowledge distillation to pruned student (ablation) | KDIC · CLIC | BD-rate (%): -10.05 -> -12.63 | Improved |
| add knowledge distillation to pruned student (ablation) | KDIC · Tecnick | BD-rate (%): -13.92 -> -16.64 | Improved |
| prune teacher into student (40% fewer parameters, 57% fewer FLOPs) (ablation) | KDIC · Kodak | BD-rate (%): -14.28 -> -13.71 | No clear effect |
| use stage-wise implicit supervision instead of end-to-end RD + MSE feature loss (ablation) | KDIC · Kodak | BD-rate (%): -11.22 -> -13.91 | Improved |
| use finer six-stage split instead of two-stage encoder/decoder implicit supervision (ablation) | KDIC · Kodak | BD-rate (%): -12.74 -> -13.91 | Improved |
| use implicit end-to-end RD supervision instead of explicit MSE feature supervision (ablation) | KDIC · Kodak | BD-rate (%): -12.07 -> -13.91 | Improved |
| align teacher and student architecture (hybrid teacher for hybrid student) (ablation) | KDIC · Kodak | BD-rate (%): -11.01 -> -13.71 | Improved |
| add distillation to Conv-KD student (ablation) | Conv-KD (S2C-Conv) · Kodak | BD-rate (%): -8.77 -> -11.49 | Improved |
| add distillation to TCM-KD student (ablation) | TCM-KD (TCM-L) · Kodak | BD-rate (%): -7.49 -> -9.79 | Improved |
| KDIC versus MLIC++ and TCM-L complexity | KDIC · 2K images | parameters: 116.48 -> 47.61 | Improved |
Closest papers by meaning
- When the Teacher Has More Bits: Self-Teacher Latent Distillation for Learned Image Compression (ECCV 2026 · similarity 0.92)
- Learning to Explore Distillability and Sparsability: A Joint Framework for Model Compression (TPAMI 2022 · similarity 0.89)
- Knowledge Distillation: A Survey (IJCV 2021 · similarity 0.89)
- EPSD: Early Pruning with Self-Distillation for Efficient Model Compression (AAAI 2024 · similarity 0.87)
- Knowledge Distillation with Multi-granularity Mixture of Priors for Image Super-Resolution (ICLR 2025 · similarity 0.87)
- Few-Shot Domain Adaptation for Learned Image Compression (AAAI 2025 · similarity 0.86)
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.
