Research map / Image generation / Image restoration
COMPASS: High-Efficiency Deep Image Compression with Arbitrary-scale Spatial Scalability
Jongmin Park, Jooyoung Lee, Munchurl Kim
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 |
|---|---|---|---|
| replace bicubic inter-layer prediction with LIFF (ablation) | COMPASS · Kodak, 3-layer | BD-rate vs full model: 9.27% BD-rate loss without LIFF | Improved |
| add convolutional-layer-wise padding instead of padding to multiple of 64 (ablation) | COMPASS · Kodak, 3-layer | BD-rate vs full model: 18.95% BD-rate loss without it | Improved |
| use rounded latent instead of noisy latent in the distortion term (ablation) | COMPASS · Kodak, 3-layer | BD-rate vs full model: 13.00% BD-rate loss without it | Improved |
| use COMPASS with arbitrary-scale LIFF instead of fixed-scale Mei et al. enhanced | spatially scalable coding · Kodak, 3-layer, scale 2 | BD-rate: -14.23% | Improved |
| use COMPASS instead of single-layer coding at scale 1.2x | spatially scalable coding · Kodak, 2-layer | BD-rate: -8.19% | Improved |
| use COMPASS instead of single-layer coding at scale 2.0x | spatially scalable coding · Kodak, 2-layer | BD-rate: 8.80% | Hurt |
| use COMPASS instead of single-layer coding at scale 4.0x | spatially scalable coding · Kodak, 3-layer | BD-rate: 4.74% | Hurt |
| use shared LIFF and residual modules across enhancement layers | COMPASS · Kodak, 3-layer | parameters: 15.5M vs 40.7M (Mei et al.) | Improved |
Closest papers by meaning
- Learning End-to-End Lossy Image Compression: A Benchmark (TPAMI 2021 · similarity 0.88)
- Neural Image Compression: Generalization, Robustness, and Spectral Biases (NeurIPS 2023 · similarity 0.88)
- Neural Image Compression via Attentional Multi-Scale Back Projection and Frequency Decomposition (ICCV 2021 · similarity 0.87)
- Computationally-Efficient Neural Image Compression with Shallow Decoders (ICCV 2023 · similarity 0.87)
- Enhanced Neural Video Representation Compression with High Scalability (ECCV 2026 · similarity 0.86)
- Towards Practical Real-Time Neural Video Compression (CVPR 2025 · similarity 0.84)
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.
