AI Paper Atlas

Research map / Image generation / Image restoration

Joint Demosaicking and Denoising by Fine-Tuning of Bursts of Raw Images

Thibaud Ehret, Axel Davy, Pablo Arias, Gabriele Facciolo

ICCV 2019 · Image generation › Image restoration · approach: Raw and low-light noise modeling · 71 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train without RGB ground truth using mosaic-to-mosaic loss instead of supervised ground truthSection 2 demosaicking network · Kodak demosaickingPSNR (dB): 41.2 -> 41.3Improved
fine-tune on a single image's burst (ablation)Section 2 network · Kodak kodim19PSNR (dB): 42.1 -> 44.4Improved
fine-tune on kodim19 and evaluate on whole dataset (ablation)Section 2 network · Kodak (full dataset)PSNR (dB): 41.3 -> 40.4Hurt
add fine-tuning to Gharbi et al. networkGharbi et al. · Kodak all sigma=5PSNR (dB): 36.2 -> 37.6Improved
add fine-tuning to Gharbi et al. networkGharbi et al. · Kodak 19 sigma=10PSNR (dB): 33.2 -> 34.0Improved
fine-tune the noise-free Section 2 network (never saw noise)Section 2 network · Kodak all sigma=5PSNR (dB): 36.0 vs Kokkinos 35.2, Gharbi 36.2Mixed
lengthen burst from 10 to 20 images (ablation)Section 2 network + fine-tuning · Kodak 19 sigma=10PSNR (dB): 32.9 -> 33.2Improved
lengthen burst from 10 to 20 images (ablation)Gharbi et al. + fine-tuning · Kodak 19 sigma=10PSNR (dB): 34.0 -> 34.3Improved
use L1 instead of L2 loss for fine-tuning on clipped noise (ablation)DnCNN fine-tuned on burst · saturated-region grayscale burstPSNRImproved
fine-tune DnCNN on a burstDnCNN sigma=25 · 10-frame noisy grayscale burstPSNRImproved
fine-tune on a self-similar stripes image vs a binary noise image (ablation)DnCNN fine-tuning · synthetic stripes vs binary noisePSNRMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy