Densifying Sparse Ocean Depth Observations

DepthDif explores conditional diffusion for reconstructing dense subsurface ocean temperature fields from sparse, masked observations.

The repository currently supports: - single-band corrupted-to-clean reconstruction - EO-conditioned multi-band reconstruction (surface condition + deeper target bands)

Model Description

DepthDif is a conditional diffusion model: it reconstructs dense depth fields from corrupted submarine observations, conditioned on EO (surface) data plus sparse corrupted subsurface input. It can inject coordinate/date context via FiLM conditioning and reconstruct the full target image. See the full model details in Model.

Documentation Map

• Quick Start: environment setup + fastest train/infer path
• Data: dataset source, export format, masking pipeline, split behavior
• Model: architecture and diffusion conditioning flow
• Date + Coordination Injection: coordinate/date FiLM conditioning details
• Training: CLI usage, run outputs, logging, checkpoints
• Inference: script and direct predict_step workflows
• Sampling Diagnostics: denoising intermediates, MAE-vs-step, and schedule profiling
• Experiments: qualitative test results
• Model Settings: key config knobs and where they are used
• Development: known issues, TODOs, and roadmap
• API Reference: auto-generated module reference via mkdocstrings