Zero-shot diffusion image compression with sectioned codebooks and one channel code
Dinanath Padhya
Abstract
Zero-shot diffusion codecs such as Turbo-DDCM compress images without any training. In each coded step, the encoder selects noise vectors from one codebook and transmits a single index for the selected set. We propose sectioned codebooks. A coded step is divided into sections, and each section has its own large codebook and transmits its best vector as a short field of bits. A model based on order statistics quantifies how well the bits of a step are used and predicts the gain of sectioned codebooks from measurements of Turbo-DDCM alone. Fast Walsh-Hadamard transforms make sections of 220 vectors practical. At equal bits, sectioned codebooks improve the PSNR of Turbo-DDCM by 0.18 to 0.51 dB on three datasets, improve LPIPS and DISTS, and save 16 to 21% of the bits at equal PSNR. Codecs that are robust to channel errors by design are usually compared only with unprotected codecs. We compare them, at equal total bits, with a codec protected by one channel code. With one BCH code, the sectioned codec outperforms all reported results of Robust Turbo-DDCM and BitFlat on two datasets, by up to 7.74 dB at the highest bit error rate. At the lowest bit error rate its margin over BitFlat is only 0.08 to 0.16 dB. When the code fails, the receiver can still decode the fields as received. Additional experiments cover burst errors and packet loss, and all observed gains of robust codecs stay within a limit derived from the channel capacity.
Create a lesson
Related papers
AnyVIO: Feed-Forward Streaming Visual-Inertial Odometry Across Platforms
Omoruyi Atekha, Marcus Greiff, Ray Zhang et al.
Susceptibility source separation via Closed-form Alternating Newton-type Optimization
Ignacio Contreras-Zúñiga, Cristian Tejos, Carlos Milovic
Physics-Informed Deep Learning for Shear Wave Speed Estimation in MR Elastography
Stefan Martin, Jakob Schattenfroh, Patrick Schuenke et al.
LBW-NPS: A Local Band-Weighted Noise Power Spectrum Loss for Texture-Preserving Low-Dose CT Denoising
Fatma Elif Yalçındağ, Mehmet Ozan Ünal, Metin Ertaş et al.
Life-IQA+: Boosting Blind Image Quality Assessment via GCN-enhanced Layer Interaction and Flexible Hierarchical Expert Specialization
Long Tang, Guoquan Zheng, Huiyu Duan et al.
Uncertainty-Aware Multimodal Deep Learning for Prediction of Radiation-Induced Xerostomia in Head and Neck Cancer
Jin Yang, Tian Liu, Jose Penagaricano et al.