The paper introduces NOISE2VOID (N2V), a novel training scheme for image denoising that does not require clean target images or noisy image pairs. Unlike traditional methods that rely on pairs of noisy and clean images, N2V trains directly on a body of noisy images, making it applicable to scenarios where clean targets are unavailable, such as biomedical image data. The authors demonstrate the effectiveness of N2V by comparing its performance to other denoising methods on various datasets, including the BSD68 dataset and simulated microscopy data. While N2V does not outperform methods with more information during training, it shows moderate performance degradation and still outperforms some self-supervised methods like BM3D and non-local means. The paper also discusses the limitations of N2V, particularly when the assumptions of predictable signal and pixel-wise independent noise are violated, and provides a detailed implementation and experimental results.The paper introduces NOISE2VOID (N2V), a novel training scheme for image denoising that does not require clean target images or noisy image pairs. Unlike traditional methods that rely on pairs of noisy and clean images, N2V trains directly on a body of noisy images, making it applicable to scenarios where clean targets are unavailable, such as biomedical image data. The authors demonstrate the effectiveness of N2V by comparing its performance to other denoising methods on various datasets, including the BSD68 dataset and simulated microscopy data. While N2V does not outperform methods with more information during training, it shows moderate performance degradation and still outperforms some self-supervised methods like BM3D and non-local means. The paper also discusses the limitations of N2V, particularly when the assumptions of predictable signal and pixel-wise independent noise are violated, and provides a detailed implementation and experimental results.