Xiao-Yue Ling

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. We present PrismCodec, which recasts a pretrained Diffusion Transformer (DiT) as a bitstream-grounded one-ste...