| """Build data/metadata.csv and scripts/cache/clusters.csv from raw ViViD-5k data. |
| |
| Loads the 515 MB COCO json once (~1-2 min). Fails loudly on any mismatch |
| between images, point files, split entries, and COCO entries. |
| """ |
| import json |
| import sys |
| from pathlib import Path |
|
|
| import numpy as np |
| import pandas as pd |
| from tqdm import tqdm |
|
|
| sys.path.insert(0, str(Path(__file__).resolve().parent)) |
| from vivid_utils import load_splits, parse_bbox |
|
|
| ROOT = Path(__file__).resolve().parents[1] |
| DATA = ROOT / "data" |
| CACHE = ROOT / "scripts" / "cache" |
| GRAPE_CATEGORY_ID = 1 |
|
|
|
|
| def main(): |
| splits = load_splits(DATA) |
| img_files = {p.name: p for p in (DATA / "imgs").iterdir()} |
| point_files = {p.stem: p for p in (DATA / "anns" / "points").glob("*.npy")} |
|
|
| errors = [] |
| for name in splits: |
| if name not in img_files: |
| errors.append(f"in split but no image file: {name}") |
| if Path(name).stem not in point_files: |
| errors.append(f"no points npy for image: {name}") |
| for name in img_files: |
| if name not in splits: |
| errors.append(f"image not in any split: {name}") |
| if errors: |
| print("\n".join(errors), file=sys.stderr) |
| sys.exit(f"ABORT: {len(errors)} consistency errors") |
|
|
| print("Loading COCO json (this takes a minute)...") |
| coco = json.loads((DATA / "anns" / "instances_updated.json").read_text()) |
| coco_by_id = {im["id"]: im for im in coco["images"]} |
| coco_by_name = {im["file_name"]: im for im in coco["images"]} |
|
|
| missing_coco = [n for n in splits if n not in coco_by_name] |
| if missing_coco: |
| sys.exit(f"ABORT: {len(missing_coco)} images missing from COCO json: {missing_coco[:5]}") |
|
|
| cluster_count = {n: 0 for n in splits} |
| cluster_rows = [] |
| skipped_ids = [] |
| for ann in tqdm(coco["annotations"], desc="annotations"): |
| if ann["category_id"] != GRAPE_CATEGORY_ID: |
| continue |
| if not ann["bbox"]: |
| skipped_ids.append(ann["id"]) |
| continue |
| im = coco_by_id[ann["image_id"]] |
| name = im["file_name"] |
| cluster_count[name] += 1 |
| x, y, w, h = parse_bbox(ann["bbox"]) |
| cluster_rows.append( |
| {"file_name": name, "bbox_w": w, "bbox_h": h, |
| "rel_area": (w * h) / (im["width"] * im["height"])} |
| ) |
| if skipped_ids: |
| print(f"WARNING: skipped {len(skipped_ids)} degenerate annotations " |
| f"(empty bbox): ids={sorted(skipped_ids)}") |
|
|
| rows = [] |
| for name, split in tqdm(sorted(splits.items()), desc="images"): |
| pts = np.load(point_files[Path(name).stem]) |
| if pts.ndim != 2 or pts.shape[1] != 2: |
| sys.exit(f"ABORT: unexpected points shape {pts.shape} for {name}") |
| im = coco_by_name[name] |
| rows.append( |
| {"file_name": name, "split": split, "berry_count": len(pts), |
| "cluster_count": cluster_count[name], |
| "width": im["width"], "height": im["height"]} |
| ) |
|
|
| meta = pd.DataFrame(rows) |
| meta.to_csv(DATA / "metadata.csv", index=False) |
| CACHE.mkdir(parents=True, exist_ok=True) |
| pd.DataFrame(cluster_rows).to_csv(CACHE / "clusters.csv", index=False) |
|
|
| print(f"TOTALS images={len(meta)} berries={meta.berry_count.sum()} " |
| f"clusters={meta.cluster_count.sum()}") |
| print(meta.groupby("split").size()) |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
