Research map / Image generation / Image restoration
FreqSIC: Frequency-aware Stereo Image Compression with Bi-directional Checkerboard Context Model
Shiyu Qin, Yongkang Lu, Yimin Zhou, Jiawei Li, Yifan Ren, Yuerong Xue, Shu-Tao Xia, Bin Chen
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 |
|---|---|---|---|
| use FSCT instead of plain stereo cross attention (ablation) | FreqSIC · InStereo2K | BD-PSNR vs full model: -0.251 dB without it | Improved |
| add frequency dynamic fusion (adaptive weighting of high/low bands) (ablation) | FreqSIC FSCT · InStereo2K | BD-PSNR vs full model: -0.178 dB without it | Improved |
| use learnable pooling-based frequency decomposition instead of Haar wavelet (ablation) | FreqSIC FSCT · InStereo2K | BD-PSNR vs full model: -0.906 dB with Haar | Improved |
| use FSCT instead of BiSIC mutual attention (ablation) | FreqSIC · Cityscapes | BD-PSNR vs full model: -0.203 dB with mutual attention | Improved |
| add inter-view priors in the checkerboard context model (ablation) | FreqSIC · InStereo2K | BD-PSNR vs full model: -0.648 dB without them | Improved |
| use bi-directional checkerboard context model instead of BCSIC context model (ablation) | FreqSIC · Cityscapes | BD-PSNR vs full model: -0.648 dB with BCSIC | Improved |
| use checkerboard context model instead of BiSIC autoregressive context model (ablation) | FreqSIC · InStereo2K | BD-PSNR vs full model: -0.135 dB with BiSIC | Improved |
| replace spatial autoregressive entropy model with checkerboard context model | FreqSIC vs BiSIC · InStereo2K coding latency | total latency (s): 78.60 -> 1.62 | Improved |
| FreqSIC versus BiSIC overall | FreqSIC · InStereo2K | BD-Rate (PSNR): -48.07% -> -52.43% | Improved |
| increase target bitrate (lambda) in FDF module | FreqSIC · InStereo2K | high-frequency weight | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CAMSIC — CAMSIC: Content-aware Masked Image Modeling Transformer for Stereo Image Compression (AAAI 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MambaSIC: Mamba-based Stereo Image Compression with Bi-directional Multi-reference Entropy Model (CVPR 2026 · similarity 0.90)
- CAMSIC: Content-aware Masked Image Modeling Transformer for Stereo Image Compression (AAAI 2025 · similarity 0.89)
- 3D-LMVIC: Learning-based Multi-View Image Compression with 3D Gaussian Geometric Priors (ICML 2025 · similarity 0.87)
- Low-Latency Neural Stereo Streaming (CVPR 2024 · similarity 0.86)
- Adaptive Multi-Modal Cross-Entropy Loss for Stereo Matching (CVPR 2024 · similarity 0.85)
- CAT-3DGS: A Context-Adaptive Triplane Approach to Rate-Distortion-Optimized 3DGS Compression (ICLR 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.
