What it does

Awesome-Denoise is a living bibliography of deep-learning denoising papers from top-tier venues (CVPR, ICCV, NeurIPS, TIP). Each entry includes publication year, citation count, PDF link, and code repository when available. Papers are cross-indexed by color space (RGB, raw, or both), input type (single image, burst, video), and noise model—AWGN, Poisson-Gaussian, GAN-based, or real sensor noise.

The interesting bit

The taxonomy is the work. Most “awesome” lists dump papers chronologically; this one sorts by whether you’re fighting synthetic Gaussian noise or the actual garbage a smartphone sensor produces at midnight. The README also tracks lineage—how video denoising papers quietly downgrade to burst or single-image tasks, and how raw-domain work sometimes sneaks through an ISP to sRGB.

Key highlights

• Covers self-supervised methods (Noise2Noise, Noise2Void, and their ever-growing progeny) plus supervised and traditional approaches
• Benchmark datasets catalogued: SIDD, RENOIR, PolyU, SID, DND with citation counts and (mostly working) code links
• Papers tagged by practical axes: color space, image kind, noise model—so you can find “real-world raw burst denoising” without reading 300 abstracts
• Includes 2023 papers with 0–1 citations, which is either thoroughness or optimism

Caveats

• Some dataset links are broken (RENOIR’s homepage is noted as dead)
• README is truncated mid-table in the source; full coverage of 2021–2023 papers is unclear
• No code or tools in this repo itself—it’s pure curation, not a framework

Verdict

Grab this if you’re writing a literature review, choosing a baseline for your denoising paper, or trying to prove your method isn’t just Noise2Noise with a fancier acronym. Skip it if you want runnable code or theoretical depth—the value is navigation, not education.