Residual Contrastive Learning for Image Reconstruction: Learning Transferable Representations from Noisy Images
This work addresses the challenge of reducing annotation costs for image reconstruction tasks, offering a domain-specific solution that is incremental by building on contrastive learning with a novel residual-based approach.
The paper tackles the problem of learning transferable representations for low-level image reconstruction tasks from noisy images without extensive labeled data, proposing Residual Contrastive Learning (RCL) to align pretext tasks with downstream goals, resulting in improved performance on tasks like denoising and super-resolution compared to existing self-supervised methods.
This paper is concerned with contrastive learning (CL) for low-level image restoration and enhancement tasks. We propose a new label-efficient learning paradigm based on residuals, residual contrastive learning (RCL), and derive an unsupervised visual representation learning framework, suitable for low-level vision tasks with noisy inputs. While supervised image reconstruction aims to minimize residual terms directly, RCL alternatively builds a connection between residuals and CL by defining a novel instance discrimination pretext task, using residuals as the discriminative feature. Our formulation mitigates the severe task misalignment between instance discrimination pretext tasks and downstream image reconstruction tasks, present in existing CL frameworks. Experimentally, we find that RCL can learn robust and transferable representations that improve the performance of various downstream tasks, such as denoising and super resolution, in comparison with recent self-supervised methods designed specifically for noisy inputs. Additionally, our unsupervised pre-training can significantly reduce annotation costs whilst maintaining performance competitive with fully-supervised image reconstruction.