Recasting DiT for Extreme Image Compression with Hierarchical Visual Priors
Pretrained diffusion models provide powerful generative priors for extreme image compression, but their stochastic, iterative generation processes are not designed to decode compressed representations directly.
ProofPaper ↗
Key points
- We present PrismCodec, which recasts a pretrained Diffusion Transformer (DiT) as a bitstream-grounded one-step generative decoder.
- PrismCodec uses the decoded representation in two complementary roles: a projected codec state defines the starting state for generation, while hierarchical visual priors derived from the same decoded latent provide semantic and structural guidance.
- To bridge the gap between stochastic, iterative generation and deterministic one-step decoding, we develop a progressive adaptation strategy.
- Across Kodak, CLIC, and DIV2K, PrismCodec consistently improves rate-perception performance over prior generative codecs, achieving BD-rate reductions of 70-75% in LPIPS and 75-81% in DISTS over StableCodec.
Sources (1)
- [1]Recasting DiT for Extreme Image Compression with Hierarchical Visual PriorsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 03:58 PM
Pretrained diffusion models provide powerful generative priors for extreme image compression, but their stochastic, iterative generation processes are not designed to decode compressed representations directly.
We present PrismCodec, which recasts a pretrained Diffusion Transformer (DiT) as a bitstream-grounded one-step generative decoder.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 13, 2026LinMorph: Learning to Morph Weights for Efficient Video DiT Linearization
- Oct 13, 2026Risk-Averse Inference-Time Diffusion Model Alignment with Finite-Sample Guarantees
- Oct 13, 2026Diff Diffusion: Predicting Edit Paths with Diffusion Models for Efficient Code Editing
- Oct 13, 2026FlashVSR+: Efficient Video Super-Resolution via Hybrid Attention and Reward-Yield Distribution Matching
- Oct 12, 2026Image Tokenizers Should Be Trained Between Images
- Oct 12, 2026ProMoDD: Dataset Distillation for 3D Human Motion via Prototype-Conditioned Diffusion Models