BitFlat: Error-Localized Bitstream Design for Robust Diffusion Image Compression
Zero-shot diffusion image compression based on reverse channel coding achieves high perceptual quality at low bitrates without retraining, yet its bitstream remains vulnerable to bit flips.
ProofPaper ↗
Key points
- To address these issues, we propose BitFlat, a training-free error-localizing bitstream framework with two components.
- Blockwise Locality Alignment (BLA) decomposes the globally coupled atom selection at each diffusion step into independent selection units, each describing one signed atom from a dedicated codebook with a fixed-width field, so that any bit error affects only the unit in which it occurs.
- Reliability-Aware Diffusion Control (RADC) uses the bits saved by BLA to impose structured valid-codeword constraints within each field and converts the decoding state of each field into a binary reliability gate, which turns detected corrupted atoms into erasures at known positions and excludes them from noise synthesis.
- Experiments on Kodak, DIV2K, and CLIC2020 show that BitFlat outperforms conventional, learned, and diffusion-based codecs at bit error rates from $10^{-4}$ to $10^{-2}$ and, even with $26.7%$ fewer bits, surpasses Robust Turbo-DDCM.
Sources (1)
- [1]BitFlat: Error-Localized Bitstream Design for Robust Diffusion Image CompressionarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 02:32 AM
Zero-shot diffusion image compression based on reverse channel coding achieves high perceptual quality at low bitrates without retraining, yet its bitstream remains vulnerable to bit flips.
To address these issues, we propose BitFlat, a training-free error-localizing bitstream framework with two components.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
- Oct 18, 2026Steered Not Stirred: Effective Reward Tilting using Surrogate Reward Mixing
- Oct 18, 2026Holo4D: Holistic 4D Reconstruction as Geometric Control for Video Diffusion
- Oct 18, 2026Google Open-Sources Harness2, a Recursive Agent Harnessing Framework
- Oct 18, 2026Investing in the public benefit
- Oct 16, 2026DiffusionRVGE: Reinforcement Learning via Regression-Based Value Gradient Estimation