Reproduce “End-to-end Optimized Image Compression” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Johannes Ballé, Valero Laparra, Eero P. Simoncelli
Researchers from NYU and Universitat de València developed an end-to-end optimized image compression method utilizing learned nonlinear transforms, including Generalized Divisive Normalization. This approach achieves superior visual quality and competitive PSNR compared to JPEG and JPEG 2000 by jointly optimizing all compression stages and effectively handling the non-differentiable quantization process.
Key insights
- End-to-end optimization of image compression systems using deep learning can outperform traditional, separately optimized methods, particularly in visual quality.
- Biologically inspired non-linearities like Generalized Divisive Normalization (GDN) are effective for learning efficient image representations for compression.
- Approximating the non-differentiable quantization process with additive uniform noise allows for end-to-end gradient-based training of discrete compression systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.