Research map / Architectures and efficiency / RNNs and normalization
BaSIC: BayesNet Structure Learning for Computational Scalable Neural Image Compression
Yufeng Zhang, Hang Yu, Shizhan Liu, Wenrui Dai, Weiyao Lin
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 |
|---|---|---|---|
| learn intra-node BayesNet structure instead of hand-crafted autoregressive structure with the same number of parallel stages (ablation) | BaSIC · Kodak (2, 4, 10 stages) | BPP: 0.3216 -> 0.3207 (2 stages vs CB); 0.3174 -> 0.3156 (10 stages vs ELIC) | Improved |
| learned 10-stage BayesNet versus non-parallel MaskConv (ablation) | BaSIC · Kodak | BPP: 0.3156 vs 0.3150 | No clear effect |
| learned BayesNet versus hand-crafted models in speed (ablation) | BaSIC · Kodak | decompression speed: 17.85 vs 17.32 MB/s (10 stages) | No clear effect |
| use inter-node BayesNet with fixed latent channels instead of SlimCAE lambda-scheduling | BaSIC · Kodak, scalable backbone | bits saved: around 2% more bits saved at max complexity, around 10% at min complexity | Improved |
| parallelize autoregressive context with fewer stages (checkerboard/channel-wise) versus MaskConv | joint autoregressive NIC · Kodak | decompression speed: 0.32 -> 345.55 MB/s (null-context baseline) | Improved |
Closest papers by meaning
- A Bayesian Optimization Framework for Neural Network Compression (ICCV 2019 · similarity 0.88)
- Neural Image Compression: Generalization, Robustness, and Spectral Biases (NeurIPS 2023 · similarity 0.86)
- Compression with Bayesian Implicit Neural Representations (NeurIPS 2023 · similarity 0.86)
- Non-vacuous Generalization Bounds at the ImageNet Scale: a PAC-Bayesian Compression Approach (ICLR 2019 · similarity 0.86)
- Robustly overfitting latents for flexible neural image compression (NeurIPS 2024 · similarity 0.85)
- BMRS: Bayesian Model Reduction for Structured Pruning (NeurIPS 2024 · 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.
