LOBSTgER-enhance: an underwater image enhancement pipeline
This addresses the challenge of heavy post-processing for underwater photographers, though it is incremental as it applies an existing diffusion method to a specific domain.
The paper tackles the problem of underwater image degradation by developing a diffusion-based pipeline that learns to reverse synthetic corruptions, achieving high perceptual consistency and strong generalization in synthesizing 512x768 images with a model of ~11M parameters trained on ~2.5k images.
Underwater photography presents significant inherent challenges including reduced contrast, spatial blur, and wavelength-dependent color distortions. These effects can obscure the vibrancy of marine life and awareness photographers in particular are often challenged with heavy post-processing pipelines to correct for these distortions. We develop an image-to-image pipeline that learns to reverse underwater degradations by introducing a synthetic corruption pipeline and learning to reverse its effects with diffusion-based generation. Training and evaluation are performed on a small high-quality dataset of awareness photography images by Keith Ellenbogen. The proposed methodology achieves high perceptual consistency and strong generalization in synthesizing 512x768 images using a model of ~11M parameters after training from scratch on ~2.5k images.