Research map / Image generation / Image restoration
DLF: Extreme Image Compression with Dual-generative Latent Fusion
Naifu Xue, Zhaoyang Jia, Jiahao Li, Bin Li, Yuan Zhang, Yan Lu
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 the detail branch (dual-branch coding) alongside the semantic branch (ablation) | DLF · CLIC2020 | BD-Rate LPIPS/DISTS: 47.9% -> 0.0% (LPIPS), 47.6% -> 0.0% (DISTS) | Improved |
| add the detail branch (ablation) | DLF · Kodak | BD-Rate LPIPS/DISTS: 17.5% -> 0.0% (LPIPS), 20.2% -> 0.0% (DISTS) | Improved |
| add cross-branch interactive transform (IT) modules (ablation) | DLF · Kodak | BD-Rate LPIPS/DISTS: 64.1% -> 0.0% (LPIPS), 73.6% -> 0.0% (DISTS) | Improved |
| add cross-branch interactive transform (IT) modules (ablation) | DLF · CLIC2020 | BD-Rate LPIPS/DISTS: 68.8% -> 0.0% (LPIPS), 61.8% -> 0.0% (DISTS) | Improved |
| use scalar quantization instead of vector quantization in detail branch (ablation) | DLF · Kodak | BD-Rate LPIPS/DISTS: 18.3% -> 0.0% (LPIPS), 40.7% -> 0.0% (DISTS) | Improved |
| use scalar quantization instead of vector quantization in detail branch (ablation) | DLF · CLIC2020 | BD-Rate LPIPS/DISTS: 27.3% -> 0.0% (LPIPS), 58.1% -> 0.0% (DISTS) | Improved |
| dual-VQ bottleneck (VQ detail branch) versus single semantic branch (ablation) | DLF · Kodak | BD-Rate LPIPS: w/ VQ detail 18.3% vs w/o detail 17.5% | No clear effect |
| DLF versus MS-ILLM (extreme low bitrate generative codec) | DLF · CLIC2020 | BD-Rate LPIPS/DISTS: -27.93% LPIPS, -53.55% DISTS | Improved |
| DLF versus diffusion codecs DiffEIC/PerCo in decoding time | DLF · Kodak | decoding time (s): 0.252 vs 2.443 (PerCo) and 4.093 (DiffEIC) | Improved |
| DLF versus MS-ILLM in coding time | DLF · Kodak | encoding/decoding time (s): 0.064 -> 0.178 enc, 0.070 -> 0.252 dec | Hurt |
Closest papers by meaning
- Generative Latent Coding for Ultra-Low Bitrate Image Compression (CVPR 2024 · similarity 0.89)
- Ultra Lowrate Image Compression with Semantic Residual Coding and Compression-aware Diffusion (ICML 2025 · similarity 0.88)
- Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion (CVPR 2025 · similarity 0.88)
- Decouple Distortion from Perception: Region Adaptive Diffusion for Extreme-low Bitrate Perception Image Compression (CVPR 2025 · similarity 0.87)
- Towards image compression with perfect realism at ultra-low bitrates (ICLR 2024 · similarity 0.87)
- One-Step Diffusion-Based Image Compression with Semantic Distillation (NeurIPS 2025 · similarity 0.87)
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.
