Search papers, labs, and topics across Lattice.
To overcome the prohibitively slow per-instance optimization of implicit neural representations (INRs), the authors develop PIC, a feedforward INR image codec that predicts all representation parameters in a single forward pass. This addresses the long-standing deployment bottleneck of neural codecs, which have historically struggled to balance rate-distortion efficiency with real-time operational constraints. Evaluated against traditional baselines, PIC achieves an encoding throughput of 20 FPS and an unprecedented 2000 FPS sub-millisecond decoding speed while matching or outperforming JPEG across rate-distortion metrics.
Neural image compression can finally rival classic formats in raw throughput, achieving 2000 FPS decoding and practical 20 FPS encoding while matching JPEG's rate-distortion performance.
Implicit neural representation (INR) has achieved remarkable progress in novel view synthesis and image/video coding in recent years.Compared to conventional end-to-end image codecs, INR-based compressors demonstrate significant advantages in decoding complexity. However, their practical application has been hindered by the inferior encoding speed and underutilized decoding efficiency.In this work, we propose a feedforward INR image coding architecture, Practical INR Image Codec (PIC), that computes all the necessary information for INR network in a single forward pass, achieving an encoding speed of 20 FPS. Additionally, we implement a highly optimized decoder that reaches 2000 FPS decoding speed, significantly surpassing JPEG's performance at comparable rate-distortion (RD) performance. To the best of our knowledge, this work presents the first learning-based image codec that simultaneously outperforms or is comparable with JPEG in both RD performance and decoding speed while maintaining practical encoding speed. Code is available at https://github.com/actcwlf/PIC.
Tsinghua AI