Image compression has advanced for more than three decades through international standards such as JPEG, JPEG2000, and HEVC, forming a technological foundation for broadcasting, communications, and internet services. Conventional approaches were based on transform coding theory and carefully designed probabilistic models. With the rapid progress of deep learning, however, a new paradigm has emerged in which signal representations and entropy models can be jointly optimized in an end-to-end manner.

The recipients were among the first to systematically introduce deep neural networks into image coding and to establish the field now known as learned image compression. They proposed an end-to-end framework integrating nonlinear analysis and synthesis transformation based on deep autoencoders with principal component analysis for latent decorrelation (1). By introducing discretized Gaussian mixture likelihoods and attention mechanisms (2), they significantly improved probability estimation in latent space and achieved compression efficiency surpassing established international standards. Furthermore, by incorporating hybrid Transformer–CNN architectures (3), they demonstrated state-of-the-art performance and clarified key architectural principles for neural compression. These contributions, presented at premier conferences such as CVPR, have been widely cited and have strongly influenced subsequent research.

Importantly, the recipients addressed one of the major challenges of learned compression, namely, its computational complexity. Through fixed-point arithmetic implementation and integer-oriented design (4), together with nonlinear quantization strategies, network pruning, and architectural optimization, they substantially reduced computational requirements. As a result, they achieved the world’s first real-time FPGA implementation of learned image compression, providing concrete evidence of its practical feasibility and accelerating research toward lightweight and deployable neural codecs.

Their work also extended to neural video compression (5), broadening the applicability of learned coding to spatio-temporal signals. In addition, their research aligns with and advances theoretical developments such as the hyper-prior model (6), joint autoregressive and hyper-prior modeling (7), and high-fidelity generative compression (8), contributing to the consolidation of the theoretical framework of neural compression.

A particularly significant aspect of their achievements is their direct impact on international standardization. Within ISO/IEC’s JPEG-AI initiative, their methods have been referenced as baseline models (9), thereby influencing the direction of AI-based image coding standards. This reflects not only academic excellence but also tangible global technological impact.

Through sustained contributions spanning theoretical innovation (1)–(3), practical implementation (4), application expansion (5), and standardization influence (9), the recipients have fundamentally advanced image compression technology and established a new academic domain at the intersection of information compression and artificial intelligence. Their achievements are fully deserving of the Achievement Award.

References

1. Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “A Deep Convolutional AutoEncoder-based Lossy Image Compression,” Proc. PCS, pp. 1-5, June 2018.
2. Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention Modules,” Proc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), pp. 7939-7948, June 2020.
3. J. Liu, H. Sun, and J. Katto, “Learned Image Compression with Mixed Transformer-CNN Architectures,” Proc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), pp. 14388-14397, June 2023.
4. H. Sun, L. Yu, and J. Katto, “Learned Image Compression with Fixed-point Arithmetic,” Proc. PCS, pp. 1-5, June 2021.
5. C. Zhang, H. Sun, and J. Katto, “Neural Video Compression with Learned Motion Representation,” Proc. IEEE/CVF Conf. on Computer Vision and Pattern Recognition (CVPR), pp. 12345-12354, June 2025.
6. J. Ballé, D. Minnen, S. Singh, S. J. Hwang, and N. Johnston, “Variational Image Compression with a Scale Hyperprior,” Proc. ICLR, 2018.
7. D. Minnen, J. Ballé, and G. Toderici, “Joint Autoregressive and Hierarchical Priors for Learned Image Compression,” Proc. NeurIPS, pp. 10771-10780, 2018.
8. F. Mentzer, G. Toderici, M. Tschannen, and E. Agustsson, “High-Fidelity Generative Image Compression,” Proc. CVPR, pp. 11913-11922, 2020.
9. ISO/IEC JTC1/SC29/WG1, “JPEG AI Committee Draft,” ISO/IEC CD 6048-1, 2025.