H-DDCM: Hadamard Codebooks for Fast
Generative Image and Video Compression
Abstract
Training-free diffusion-based compression methods repurpose the generative priors of a diffusion or a flow model to compress data without retraining them as task-specific codecs. Recent codebook-based approaches such as Turbo-DDCM achieve strong perceptual quality at ultra-low bitrates, but their reliance on dense Gaussian codebooks remains costly in both memory and computation, and scales poorly with the large latent dimensions of modern image and video generators. Their combinatorial bitstream encoding further becomes computationally heavy at this scale. We introduce H-DDCM, a scalable generative compression framework that replaces Gaussian codebooks with implicit randomized Hadamard dictionaries. The fast Walsh–Hadamard transform allows for implicitly synthesizing full-dimensional dictionaries with O(D log D) compute for projection and reconstruction and O(D) memory, without materializing the full-dimensional dictionary. We further introduce an efficient table-free lexicographic coding scheme that makes combinatorial encoding practical at larger dimensions. Beyond images, we adapt the framework to an autoregressive video flow model and develop a streaming compression scheme that periodically re-encodes recently reconstructed frames into a fresh KV-cache prefix, preserving temporal context across segments at no additional transmitted bits. This enables long-form, close to real-time generative video compression on a single GPU. Across image and video benchmarks, H-DDCM achieves competitive rate-distortion-perception performance while substantially reducing the memory and computational overhead of codebook-based training-free generative compression.
We compare our reproducible codebook-based noise selection against Turbo-DDCM on two pretrained diffusion backbones, Stable Diffusion 2.1 (Kodak24) and FLUX (DIV2K), across a wide bits-per-pixel (BPP) range. These quantitative comparisons show that our method is the fastest, without additional memory requirements, and achieves on-par results to other state-of-the-art methods on the rate-distortion-perception tradeoff.
Quantitative Comparison
Stable Diffusion 2.1 (Kodak24)
FLUX (DIV2K)
Qualitative Comparison — Stable Diffusion 2.1 (Kodak24)
Qualitative Comparison — FLUX (DIV2K)
We extend our reproducible codebook-based noise selection to video generation, using a Causal-Forcing (CF) autoregressive (AR) causal video model as the backbone, evaluated on the MCL-JCV and UVG datasets. We achieve the fastest encoding and decoding compression compared to methods leveraging the same CF backbone, leading towards real-time video compression and streaming, while achieving on-par performance on the rate-distortion-perception tradeoff.
Quantitative Comparison
Causal-Forcing (MCL-JCV / UVG)