ocean-SR-training

What problem does it solve?

Automates the end-to-end training, evaluation, visualization, and reporting of ocean super-resolution models, including robust error handling, OOM protection, and workflow orchestration, reducing manual setup and maintenance.

Core Features & Use Cases

• Orchestrates model selection, dataset handling, training, and inference across GPU setups using ocean_sr_train_start, ocean_sr_train_status, ocean_sr_train_visualize, and ocean_sr_train_report.
• Supports multiple architectures (FNO2d, SwinIR, UNet2d, SRNO, etc.) with safety checks, checkpoints, and break-glass recovery workflows.
• Predict mode with full-dataset inference and plots, while allowing ramp-back to training after evaluation.

Quick Start

Start an end-to-end ocean-SR training workflow by invoking ocean_sr_train_start with dataset_root, log_dir, model_name, dyn_vars, and scale to begin training, evaluation, visualization, and reporting.