Datasets:
Tasks:
Image-to-Image
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
ArXiv:
Tags:
flicker-banding
screen-capture
bracketed-exposure
raw-image
computational-photography
image-restoration
License:
| """Load a single Bricker sample. | |
| python scripts/load_sample.py <sample_dir> | |
| Example: | |
| python scripts/load_sample.py ../data/real/Test/scene_0000/000000_game | |
| Works for both subsets (`real/` and `syn/`); only numpy is required. | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import os | |
| import numpy as np | |
| FRAME_KEYS = ("I1", "I2", "I3", "I4", "I5") | |
| EXPOSURE = {"I1": "1/16x", "I2": "1/4x", "I3": "1x (reference)", "I4": "4x", "I5": "16x"} | |
| WHITE_LEVEL = 1023.0 # frames are stored as uint16 scaled to 2**10 - 1 | |
| def load_sample(sample_dir: str, normalize: bool = True) -> dict: | |
| """Read one sample directory. | |
| Returns a dict with: | |
| frames -- {key: uint16 array of shape (Hp, Wp, 4)} for I1..I5 | |
| gt -- float32 array of shape (Hp, Wp, 4), 1.0 = white level | |
| meta -- metadata dict (see README) | |
| raw -- optional uint16 stack of the five frames, shape (5, Hp, Wp, 4) | |
| """ | |
| frames = {k: np.load(os.path.join(sample_dir, f"raw_{k}.npy")) for k in FRAME_KEYS} | |
| gt = np.load(os.path.join(sample_dir, "raw_gt.npy")) | |
| meta = np.load(os.path.join(sample_dir, "metadata.npy"), allow_pickle=True).item() | |
| out = {"frames": frames, "gt": gt, "meta": meta} | |
| if normalize: | |
| out["raw"] = np.stack([frames[k].astype(np.float32) / WHITE_LEVEL for k in FRAME_KEYS]) | |
| return out | |
| def main() -> None: | |
| ap = argparse.ArgumentParser(description=__doc__) | |
| ap.add_argument("sample_dir") | |
| args = ap.parse_args() | |
| s = load_sample(args.sample_dir) | |
| print(f"sample: {args.sample_dir}") | |
| for k in FRAME_KEYS: | |
| a = s["frames"][k] | |
| print(f" raw_{k}.npy {str(a.dtype):7s} {a.shape} exposure={EXPOSURE[k]:14s} " | |
| f"min={a.min():5d} max={a.max():5d} mean={a.mean():8.1f}") | |
| print(f" raw_gt.npy {str(s['gt'].dtype):7s} {s['gt'].shape} " | |
| f"min={s['gt'].min():.3f} max={s['gt'].max():.3f}") | |
| print(f" metadata keys: {sorted(s['meta'].keys())}") | |
| print(f" CFA pattern: {s['meta']['CFAPattern']} (packing: R, G1, G2, B)") | |
| print(f" stacked frames: {s['raw'].shape} float32 in [0, 1]") | |
| if __name__ == "__main__": | |
| main() | |