| """Evaluate SatMAE++ masked and multi-scale reconstruction.""" |
| import json |
| from pathlib import Path |
| import matplotlib.pyplot as plt |
| import numpy as np, yaml |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
|
|
|
|
| def validate_config(cfg): |
| data, model = cfg["data"], cfg["model"] |
| if data["image_size"] != model["image_size"] or data["channels"] != model["in_channels"]: |
| raise ValueError("data and model image shape settings must match") |
| if data["scales"] != model["scales"]: |
| raise ValueError("data.scales and model.scales must match") |
|
|
|
|
| def patchify(images, patch_size): |
| batch, channels, height, width = images.shape |
| patches = images.reshape(batch, channels, height // patch_size, patch_size, |
| width // patch_size, patch_size) |
| return patches.transpose(0, 2, 4, 1, 3, 5).reshape(batch, -1, channels * patch_size**2) |
|
|
|
|
| def main(): |
| import argparse |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument("--config", type=Path, default=ROOT / "conf/config.yaml") |
| args = parser.parse_args() |
| cfg = yaml.safe_load(args.config.read_text()) |
| validate_config(cfg) |
| source = ROOT / cfg["paths"]["inference_dir"] / "reconstruction.npz" |
| if not source.exists(): raise FileNotFoundError("Run inference before evaluation") |
| archive = np.load(source); target, prediction = archive["target"], archive["prediction"] |
| mse = float(np.mean((prediction - target) ** 2)); masked = archive["mask"].astype(bool) |
| patch_error = np.mean((patchify(prediction, cfg["model"]["patch_size"]) - |
| patchify(target, cfg["model"]["patch_size"])) ** 2, axis=-1) |
| if masked.size != patch_error.shape[0] * patch_error.shape[1]: |
| groups = masked.reshape(masked.shape[0], -1, patch_error.shape[1]) |
| masked = groups.any(axis=1) |
| masked_mse = float(patch_error[masked].mean()) if masked.any() else mse |
| result = {"reconstruction_mse": mse, "masked_mse": masked_mse, |
| "data_source": "synthetic", "protocol": cfg["data"]["protocol"]} |
| out = ROOT / cfg["paths"]["evaluation_dir"]; out.mkdir(parents=True, exist_ok=True) |
| scales, scale_errors = cfg["model"]["scales"], [] |
| for scale in scales: |
| value = archive[f"prediction_{scale}"] |
| scaled = archive[f"target_{scale}"] |
| scale_errors.append(float(np.mean((value - scaled) ** 2))) |
| result["scale_mse"] = {str(s): e for s, e in zip(scales, scale_errors)} |
| (out / "metrics.json").write_text(json.dumps(result, indent=2) + "\n") |
| figure, axes = plt.subplots(1, len(scales) + 1, figsize=(3 * (len(scales) + 1), 3)) |
| axes[0].imshow(np.clip(target[0, :3].transpose(1, 2, 0), 0, 1)); axes[0].set_title("Original"); axes[0].axis("off") |
| for axis, scale in zip(axes[1:], scales): |
| image = archive[f"prediction_{scale}"][0] |
| axis.imshow(np.clip(image[:3].transpose(1, 2, 0), 0, 1)); axis.set_title(f"Scale x{scale}"); axis.axis("off") |
| figure.tight_layout(); figure.savefig(out / "multiscale_reconstruction.png", dpi=160); plt.close(figure) |
| figure, axis = plt.subplots(figsize=(5, 3)); axis.plot(scales, scale_errors, marker="o"); axis.set(xlabel="Reconstruction scale", ylabel="MSE", title="Multi-scale Reconstruction Comparison"); axis.grid(alpha=0.25); figure.tight_layout(); figure.savefig(out / "scale_comparison.png", dpi=160); plt.close(figure) |
| print(json.dumps(result, indent=2)); print("evaluation=", out) |
|
|
| if __name__ == "__main__": main() |
|
|
