Research map / Architectures and efficiency / Mixture of experts
MixCompress: Mixture of Experts for Variable Rate Learned Image Compression
Calvin-Khang Ta, Praneet Singh, Tong Shao, Peng Yin
Read the paperCheck 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 |
|---|---|---|---|
| add AuxT shortcuts (ablation) | Cond-Conv variable-rate LALIC · Kodak | BD-rate (PSNR) vs VTM-23.1 (%): -15.15 -> -17.06 | Improved |
| add AuxT shortcuts (ablation) | Cond-Conv variable-rate LALIC · Tecnick | BD-rate (PSNR) vs VTM-23.1 (%): -21.53 -> -23.50 | Improved |
| replace AuxT with CAT (add CAT) (ablation) | Cond-Conv + AuxT · Kodak | BD-rate (PSNR) vs VTM-23.1 (%): -17.06 -> -17.89 | Improved |
| add CAT (ablation) | Cond-Conv + AuxT · CLIC Test | BD-rate (PSNR) vs VTM-23.1 (%): -23.37 -> -24.23 | Improved |
| replace Cond-Conv layers with MoE (N=4, K=2) (ablation) | Cond-Conv + AuxT + CAT · Kodak | BD-rate (PSNR) vs VTM-23.1 (%): -17.89 -> -18.81 | Improved |
| replace Cond-Conv layers with MoE (N=4, K=2) (ablation) | Cond-Conv + AuxT + CAT · CLIC Test | BD-rate (PSNR) vs VTM-23.1 (%): -24.23 -> -26.80 | Improved |
| replace Cond-Conv layers with MoE (N=4, K=2) (ablation) | Cond-Conv + AuxT + CAT · Tecnick | BD-rate (PSNR) vs VTM-23.1 (%): -24.41 -> -27.05 | Improved |
| replace MoE with MoD (N=4, K=2) (ablation) | MixCompress · Kodak | BD-rate (PSNR) vs VTM-23.1 (%): -18.81 -> -20.12 | Improved |
| replace MoE with MoD (N=4, K=2) (ablation) | MixCompress · CLIC Valid | BD-rate (PSNR) vs VTM-23.1 (%): -24.16 -> -25.73 | Improved |
| replace MoE with MoD (N=4, K=2) (ablation) | MixCompress · Tecnick | BD-rate (PSNR) vs VTM-23.1 (%): -27.05 -> -28.88 | Improved |
| use MixCompress MoE instead of QRAF dense modulation | LALIC · Kodak | BD-rate (PSNR) vs VTM-23.1 (%): -9.32 -> -18.81 | Improved |
| use MixCompress MoD instead of Cond-Conv dense modulation | LALIC · CLIC Test | BD-rate (PSNR) vs VTM-23.1 (%): -22.05 -> -28.71 | Improved |
Closest papers by meaning
- Learned Image Compression via Sparse Attention and Adaptive Frequency (CVPR 2026 · similarity 0.88)
- When the Teacher Has More Bits: Self-Teacher Latent Distillation for Learned Image Compression (ECCV 2026 · similarity 0.87)
- On Disentangled Training for Nonlinear Transform in Learned Image Compression (ICLR 2025 · similarity 0.87)
- Once-for-All: Controllable Generative Image Compression with Dynamic Granularity Adaptation (ICLR 2025 · similarity 0.86)
- Few-Shot Domain Adaptation for Learned Image Compression (AAAI 2025 · similarity 0.86)
- Balanced Rate-Distortion Optimization in Learned Image Compression (CVPR 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.
