cs.CVApr 14, 2026

CoD-Lite: Real-Time Diffusion-Based Generative Image Compression

Authors: Zhaoyang Jia, Naifu Xue, Zihan Zheng, Jiahao Li, Bin Li, Xiaoyi Zhang, Zongyu Guo, Yuan Zhang, +2 more

Organizations: University of Science and Technology of China · Communication University of China · Microsoft Research Asia

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Recent advanced diffusion methods typically derive strong generative priors by scaling diffusion transformers. However, scaling fails to generalize when adapted for real-time compression scenarios that demand lightweight models. In this paper, we explore the design of real-time and lightweight diffusion codecs by addressing two pivotal questions. First, does diffusion pre-training benefit lightweight diffusion codecs? Through systematic analysis, we find that generation-oriented pre-training is less effective at small model scales whereas compression-oriented pre-training yields consistently better performance. Second, are transformers essential? We find that while global attention is crucial for standard generation, lightweight convolutions suffice for compression-oriented diffusion when paired with distillation. Guided by these findings, we establish a one-step lightweight convolution diffusion codec that achieves real-time 6060~FPS encoding and 4242~FPS decoding at 1080p. Further enhanced by distillation and adversarial learning, the proposed codec reduces bitrate by 85% at a comparable FID to MS-ILLM, bridging the gap between generative compression and practical real-time deployment. Codes are released at https://github.com/microsoft/GenCodec/tree/main/CoD_Lite

Figures & tables

Figure 1: Progress in generative image codecs has largely been driven by scaling models, incurring substantial decoding latency. In contrast, our codec achieves a superior trade-off between perceptual quality and coding speed, enabling real-time 10801080 p decoding on an A100 GPU while attaining near state-of-the-art FID. Decoding parameters and time are shown.
Figure 2: Analysis on diffusion pre-training at different model scales. The codecs target 0.0156 bpp on 256×256256\times 256 images.
Figure 3: Analysis on DiT in compression-oriented diffusion models. More visualizations and illustrations are in Appendix A.2 .
Figure 4: Framework overview of proposed real-time diffusion based image codec.
Figure 5: Rate-distortion curves (left) and complexity analysis (right).
Figure 6: Visual comparison with baselines. More visual results are in Appendix B.2 .
Figure 7: Ablation study via a roadmap.
Enc. / Dec. Time (ms)720×480720\times 4801920×10241920\times 10243840×21603840\times 2160
NVIDIA A100 GPU6 / 1316 / 2463 / 75
NVIDIA RTX 2080Ti GPU9 / 2146 / 47200 / 208
AMD EPYC 9V84 Processor58 / 65287 / 2021343 / 837
Table 2: Coding speed test across different resolution and devices.
ModuleEncoderDecoderLite Diff. CNNPixel Head
Parameters28.1 M10.9 M37.9 M3.2 M
kMACs / pixel3254313829
Runtime@ 1024×19201024\times 192015.5 ms2.1 ms14.5 ms6.2 ms
Table 3: Module-wise complexity break down on A100 GPU.
Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 8: Rate-perception curves on Div2K and rate-distortion curves on all datasets.
StageImage ResolutionBPP#ImagesTraining StepsBatch SizeLearning RateGPU hours (A100)
Low-Resolution Pre-Training256×256256\times 2560.0156 bpp22.1M600 K4×324\times 321×10−41\times 10^{-4}4×564\times 56
High-Resolution Pre-Training512×512512\times 5120.0039 bpp12.8 M100 K4×164\times 162×10−52\times 10^{-5}4×94\times 9
Unified Post-Training512×512512\times 5120.0039 bpp12.8 M50 K4×164\times 162×10−52\times 10^{-5}4×64\times 6
Reconstruction Fine-Tuning512×512512\times 512Target bpp12.8 M200 K4×44\times 41×10−41\times 10^{-4}4×184\times 18
Dist. & Adv. Fine-Tuning512×512512\times 512Target bpp12.8 M200 K4×84\times 81×10−51\times 10^{-5}4×434\times 43
Appendix
Table 4: Detailed configuration for each training stage. The complete pre-training pipeline requires 284284 A100 GPU hours ( ≈12\approx 12 A100 days), while fine-tuning for each target bitrate requires 244244 A100 GPU hours ( ≈10\approx 10 A100 days).
Figure 9: Rate-perception and rate-distortion curves for our large codec.
Figure 10: Additional attention map visualizations of DiT in CoD across different query locations, timesteps, and images. The observed local focus pattern is consistent across all examples.
Figure 11: More visual comparison examples.

Explore similar work

CardsList

  1. Gen2-IC: Bridging Generative Models and Image Codecs through Latent Transport

    Jun 19, 2026Yinhuan Huang, Hao Cao, Pu chen +3Learned Image CompressionGenerative Image Compression

    Tsinghua University

  2. Recasting DiT for Extreme Image Compression with Hierarchical Visual Priors

    Oct 13, 2026Yunuo Chen, Jingnan Gao, Bing He +6Learned Image CompressionGenerative Image Compression

    Shanghai Jiao Tong University · University of Sydney

  3. Generative Video Compression with Adaptive Score Distillation

    Jul 24, 2026Naifu Xue, Zhaoyang Jia, Haosen Li +7Video Diffusion ModelsDiffusion Model Distillation

    Communication University of China, Beijing, China · Microsoft Research Asia · University of Science and Technology of China, Hefei, China +1