FSID: Fully Synthetic Image Denoising via Procedural Scene Generation

TL;DR

This paper introduces FSID, a fully synthetic dataset for image denoising created through procedural scene generation, enabling effective training of denoising models without real-world data.

Contribution

The authors developed a novel Unreal engine-based synthetic data pipeline and dataset for low-level vision tasks, reducing reliance on real images.

Findings

Model trained on synthetic data achieves competitive results on real noisy images.

Synthetic dataset contains 175,000 noisy/clean image pairs.

Procedural scene generation effectively bridges the domain gap.

Abstract

For low-level computer vision and image processing ML tasks, training on large datasets is critical for generalization. However, the standard practice of relying on real-world images primarily from the Internet comes with image quality, scalability, and privacy issues, especially in commercial contexts. To address this, we have developed a procedural synthetic data generation pipeline and dataset tailored to low-level vision tasks. Our Unreal engine-based synthetic data pipeline populates large scenes algorithmically with a combination of random 3D objects, materials, and geometric transformations. Then, we calibrate the camera noise profiles to synthesize the noisy images. From this pipeline, we generated a fully synthetic image denoising dataset (FSID) which consists of 175,000 noisy/clean image pairs. We then trained and validated a CNN-based denoising model, and demonstrated that the model trained on this synthetic data alone can achieve competitive denoising results when evaluated on real-world noisy images captured with smartphone cameras.