| |
| |
| """Write Inception statistics of the real optical targets. |
| |
| Usage:: |
| |
| python scripts/make_fid_stats.py DATASET_DIR_OR_REPO OUT_DIR |
| |
| Writes ``OUT_DIR/fid_validation.npz`` and ``OUT_DIR/fid_test.npz``, computed with |
| the sen-bridge PyTorch FID so ``FID.from_stats`` can score without featurising the |
| reals. |
| """ |
|
|
| from __future__ import annotations |
|
|
| import sys |
| from pathlib import Path |
|
|
| import numpy as np |
| import torch |
| from datasets import load_dataset |
|
|
| from sen_bridge import torch as sb |
|
|
| _BATCH = 200 |
| _HALF_RANGE = 127.5 |
| _SPLITS = (("validation", "fid_validation"), ("test", "fid_test")) |
|
|
|
|
| def main() -> None: |
| """Compute and save the statistics of the validation and test targets.""" |
| data, out = sys.argv[1], Path(sys.argv[2]) |
| out.mkdir(parents=True, exist_ok=True) |
| for split, name in _SPLITS: |
| ds = load_dataset(data, split=split) |
| fid = sb.FID() |
| for start in range(0, len(ds), _BATCH): |
| stop = min(start + _BATCH, len(ds)) |
| rows = ds.select(range(start, stop)) |
| images = np.stack([np.asarray(r["optical"]) for r in rows]) |
| tensor = torch.from_numpy(images.astype(np.float32) / _HALF_RANGE - 1) |
| fid.update(tensor.permute(0, 3, 1, 2), real=True) |
| fid.save_stats(out / f"{name}.npz", real=True) |
| print(split, len(ds), "->", out / f"{name}.npz") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
