Research map / Image generation / Image restoration
Taming Hierarchical Image Coding Optimization: A Spectral Regularization Perspective
Wuyang Cong, Junqi Shi, Ming Lu, Xu Zhang, Zhan Ma
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 both spectral regularizers | DHIC-Base · Kodak/CLIC/Tecnick avg | BD-Rate (%): -11.16 -> -20.65 | Improved |
| add intra-scale frequency regularization alone (ablation) | DHIC-Base · training speed | BD-Rate / acceleration: 1.84x acceleration, BD-Rate -1.07 | Improved |
| add inter-scale latent regularization alone (ablation) | DHIC-Base · BD-Rate | BD-Rate (%): -7.66 BD-Rate, 0.91x acceleration | Improved |
| add both regularizers (ablation) | DHIC-Base · BD-Rate and speed | BD-Rate (%): 2.30x, -10.11 | Improved |
| use DWT + Conv downsampling instead of strided conv (ablation) | inter-scale regularization · BD-Rate | BD-Rate (%): -5.49 -> -7.66 | Improved |
| use L2 instead of cosine similarity (ablation) | inter-scale regularization · BD-Rate | BD-Rate (%): -6.55 -> -7.66 | Improved |
| use L2 instead of L1 (ablation) | inter-scale regularization · BD-Rate | BD-Rate (%): -7.07 -> -7.66 | Improved |
| linear 0.05 start instead of exponential schedule (ablation) | intra-scale regularization · acceleration | acceleration: 1.60x -> 1.84x | Improved |
| set delta to 0.1 rather than 0.05 or 0.2 (ablation) | inter-scale regularization · BD-Rate | BD-Rate (%): -7.15 / -8.68 -> -11.50 | Improved |
| add re-parameterization (ablation) | DHIC · Kodak BD-Rate | BD-Rate (%): -16.45 -> -19.73 | Improved |
| add FSP module (ablation) | DHIC · Kodak BD-Rate | BD-Rate (%): -19.70 -> -19.73 | No clear effect |
| add regularization to different complexity models (ablation) | DHIC · BD-Rate at 356.41 KMACs/pix | BD-Rate (%): 3.38 -> -1.74 | Improved |
Closest papers by meaning
- Learned Image Compression via Sparse Attention and Adaptive Frequency (CVPR 2026 · similarity 0.88)
- Improving the Diffusability of Autoencoders (ICML 2025 · similarity 0.86)
- Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · similarity 0.85)
- Spectral Progressive Diffusion for Efficient Image and Video Generation (NeurIPS 2026 · similarity 0.85)
- Any-Resolution AI-Generated Image Detection by Spectral Learning (CVPR 2025 · similarity 0.85)
- Learned Image Transmission with Hierarchical Variational Autoencoder (AAAI 2025 · similarity 0.85)
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.
