"""Read coco/train/labels.json, print counts per class and save one box overlay (overlay.jpg).
Run from the dataset root:  python scripts/load_example.py   (needs Pillow)"""
import collections, json
from pathlib import Path
from PIL import Image, ImageDraw

root = Path(__file__).resolve().parent.parent
d = json.loads((root / "coco/train/labels.json").read_text())
names = {c["id"]: c["name"] for c in d["categories"]}
count = collections.Counter(names[a["category_id"]] for a in d["annotations"])
print(len(d["images"]), "tiles,", len(d["annotations"]), "boxes")
for n in names.values():
    print(f"  {n:13s} {count[n]:6d} boxes")

by_img = collections.defaultdict(list)
for a in d["annotations"]:
    by_img[a["image_id"]].append(a)
img = max(d["images"], key=lambda i: len(by_img[i["id"]]))  # tile with the most boxes
im = Image.open(root / "coco/train/images" / img["file_name"]).convert("RGB")
dr = ImageDraw.Draw(im)
for a in by_img[img["id"]]:
    x, y, w, h = a["bbox"]
    dr.rectangle([x, y, x + w, y + h], outline=(255, 0, 0), width=3)
im.save(root / "overlay.jpg", quality=90)
print("saved overlay.jpg for", img["file_name"], "with", len(by_img[img["id"]]), "boxes")