Implicit Neural Representation Image Codec with Mixed Context for Fast Decoding
Keywords: Image Compression, Implicit Neural Representation, Adaptive Entropy Modeling
Abstract: Image compression using Implicit Neural Representation (INR) is an emerging technology. While it may not match the quality of cutting-edge autoencoder models, it offers two key benefits: low computational complexity and parameter-free decoding. It also surpasses many traditional and early neural compression methods in terms of quality. In this study, we introduce a new mixed auto-regressive model (MARM) to notably decrease the decoding time for the current INR codec, particularly in scenarios with limited computational resources. MARM includes our proposed auto-regressive upsampler (ARU) blocks, which are highly computationally efficient, and ARM from previous work to strike a balance between decoding time and reconstruction quality. We also suggest enhancing ARU's performance using a checkerboard two-stage decoding strategy. Moreover, the balance between quality and speed can be adjusted by the ratio of different modules. Comprehensive experiments reveal that our method significantly boosts computational efficiency while preserving image quality. It also significantly excels in decoding acceleration when the quality requirements are more lenient.
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9026
Loading