| """Step 3b: standardize released images + per-image quality statistics. |
| |
| For every image in splits.csv (split != unused, plus a quality row for every readable image): |
| - crop to the illuminated fundus field (black border removed), resize longest side to --size, JPEG q95 |
| - quality stats on the original: field-of-view fraction, mean brightness, RMS contrast, Laplacian sharpness, |
| clipped-dark / clipped-bright fractions |
| python -m fsbench.process_images --raw data/raw --bench data/bench --out data/processed --size 1024 |
| """ |
| import argparse |
| import os |
| from concurrent.futures import ProcessPoolExecutor |
| from pathlib import Path |
|
|
| import numpy as np |
| import pandas as pd |
| from PIL import Image |
|
|
| from .hashing import crop_fundus |
|
|
| Image.MAX_IMAGE_PIXELS = None |
|
|
|
|
| def _quality(img): |
| rgb = np.asarray(img.convert("RGB"), dtype=np.float32) |
| small = np.asarray(img.convert("L").resize((512, int(512 * img.height / img.width) or 1)), dtype=np.float32) |
| fov = rgb.max(axis=2) > 15 |
| g = small |
| lap = g[1:-1, 1:-1] * 4 - g[:-2, 1:-1] - g[2:, 1:-1] - g[1:-1, :-2] - g[1:-1, 2:] |
| inside = rgb[fov] if fov.any() else rgb.reshape(-1, 3) |
| lum = inside.mean(axis=1) |
| return {"fov_fraction": float(fov.mean()), "brightness": float(lum.mean()), "contrast": float(lum.std()), |
| "sharpness": float(lap.var()), "dark_clip": float((lum < 10).mean()), "bright_clip": float((lum > 245).mean()), |
| "mean_r": float(inside[:, 0].mean()), "mean_g": float(inside[:, 1].mean()), "mean_b": float(inside[:, 2].mean())} |
|
|
|
|
| def _work(args): |
| src_path, dst_path, size = args |
| try: |
| with Image.open(src_path) as im: |
| im.load() |
| im = im.convert("RGB") |
| q = _quality(im) |
| c = crop_fundus(im) |
| if max(c.size) > size: |
| s = size / max(c.size) |
| c = c.resize((max(1, round(c.width * s)), max(1, round(c.height * s))), Image.LANCZOS) |
| if dst_path: |
| Path(dst_path).parent.mkdir(parents=True, exist_ok=True) |
| c.save(dst_path, "JPEG", quality=95) |
| return {**q, "proc_width": c.width, "proc_height": c.height, "proc_error": ""} |
| except Exception as e: |
| return {"proc_error": repr(e)[:200]} |
|
|
|
|
| def main(argv=None): |
| ap = argparse.ArgumentParser() |
| ap.add_argument("--raw", required=True) |
| ap.add_argument("--bench", required=True) |
| ap.add_argument("--out", required=True) |
| ap.add_argument("--size", type=int, default=1024) |
| ap.add_argument("--workers", type=int, default=os.cpu_count()) |
| a = ap.parse_args(argv) |
| s = pd.read_csv(Path(a.bench) / "splits.csv", dtype=str, keep_default_na=False) |
| s = s[s.unused_reason != "file_missing"] |
| jobs = [(str(Path(a.raw) / r.source / r.rel_path), |
| str(Path(a.out) / r.source / f"{r.image_id}.jpg") if r.split != "unused" else "", a.size) |
| for r in s.itertuples()] |
| with ProcessPoolExecutor(a.workers) as ex: |
| res = list(ex.map(_work, jobs, chunksize=32)) |
| q = pd.DataFrame(res) |
| q.insert(0, "image_id", s.image_id.to_numpy()) |
| q.to_csv(Path(a.bench) / "quality.csv", index=False) |
| print(f"processed {len(q)} images; errors: {(q.proc_error.fillna('') != '').sum()}") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
