SNIC: Synthesized Noisy Images using Calibration
The new method reduces the PSNR gap versus real noise by 54-64% compared to standard models.
Deep Dive
Researcher Nik Bhatt developed SNIC (Synthesized Noisy Images using Calibration), a method and dataset for creating realistic image noise. It features over 6000 noisy images from four sensors, provided in both RAW and TIFF formats. When used to train denoising AI models, images synthesized with SNIC reduce the performance gap versus real noise by 54-64% compared to using manufacturer-provided noise models, enabling better AI training data.
Why It Matters
Provides higher-quality, scalable synthetic data to train more effective image denoising and enhancement AI models for photography and computer vision.