Module contrib.semseg.validation

Make some predictions for a bunch of checkpoints. See which checkpoints have the best validation loss, mean IoU, class-specific IoU, validation accuracy, and qualitative results.

Writes results to CSV files and hparam graphs (in-progress).

Functions

def load_ckpts(root: str, *, device: str = 'cpu') ‑> list[tuple[Train, torch.nn.modules.module.Module]]

Loads the latest checkpoints for each directory within root.

Arguments

root: directory containing other directories with cfg.json and model_step{step}.pt files. device: where to load models.

Returns

List of cfg, model pairs.

def main(cfg: Validation)