import json, sys
sys.argv = ["x"]
from bench import GT, norm, lev
from paddleocr import PaddleOCR
out = {}
for tier in ["PP-OCRv6_medium", "PP-OCRv6_small", "PP-OCRv6_tiny", "PP-OCRv5_server", "PP-OCRv5_mobile"]:
    ocr = PaddleOCR(text_detection_model_name=f"{tier}_det", text_recognition_model_name=f"{tier}_rec",
                    use_doc_orientation_classify=False, use_doc_unwarping=False, use_textline_orientation=False)
    out[tier] = []
    for name, gt in GT.items():
        texts = ocr.predict(f"samples/{name}")[0]["rec_texts"]
        preds = [norm(p) for p in texts]
        for g in gt:
            ng = norm(g)
            if any(ng in p for p in preds):
                continue
            best = min(preds, key=lambda p: lev(ng, p)) if preds else ""
            orig = texts[preds.index(best)] if best in preds else ""
            out[tier].append({"doc": name, "expected": g, "got": orig, "edits": lev(ng, best)})
print(json.dumps(out, ensure_ascii=False))