| """字符检测(Detection / End-to-End Spotting)任务。 |
| |
| 模型输出 JSON 数组,bbox 归一化到 0-1000,char 是该 bbox 对应的现代汉字。 |
| """ |
|
|
| from __future__ import annotations |
|
|
| import json |
| import re |
|
|
| from ._text import strip_thinking |
|
|
| PROMPT = """你是一名古文字图像检测与识别专家。 |
| |
| 任务: |
| 请检测输入图像中所有可见的古文单字符,并为每个单字符同时给出边界框 bbox 和该字符本身。 |
| |
| 检测对象: |
| - 每个独立古文字符作为一个目标 |
| - 不要检测标点、背景纹理、裂纹、装饰线、器物边缘或非文字区域 |
| - 如果一个字符残缺但仍可辨认为字符,也应检测 |
| - 如果多个笔画明显属于同一个字符,只输出一个 bbox |
| |
| 坐标要求: |
| - bbox 格式为 [x1, y1, x2, y2] |
| - x1, y1 表示左上角坐标 |
| - x2, y2 表示右下角坐标 |
| - 所有坐标必须基于整张输入图像归一化到 0–1000 的整数范围 |
| - x1 < x2,y1 < y2 |
| - 不要输出小数 |
| |
| 字符要求: |
| - char 字段填写该 bbox 区域对应的字符本身(使用现代汉字写出) |
| - 每个 bbox 只对应一个字符 |
| - 如果字符无法识别,可以使用 "[UNK]" 代替,但仍需要输出该字符的 bbox |
| |
| 排序要求: |
| - 按从上到下、从左到右的顺序排输出 |
| - idx 从 1 开始连续编号 |
| |
| 输出要求: |
| - 只输出合法 JSON |
| - 不要输出解释、Markdown、代码块或多余文字 |
| |
| JSON 输出格式: |
| [ |
| {"idx": 1, "bbox": [x1, y1, x2, y2], "char": "<该bbox对应的字符>"}, |
| {"idx": 2, "bbox": [x1, y1, x2, y2], "char": "<该bbox对应的字符>"} |
| ] |
| """ |
|
|
|
|
| def _strip_json_fence(text: str) -> str: |
| if not text: |
| return "" |
| s = text.strip() |
| m = re.match(r"^```(?:json|JSON)?\s*\n?(.*?)\n?```\s*$", s, flags=re.DOTALL) |
| return m.group(1).strip() if m else s |
|
|
|
|
| def extract(answer: str) -> tuple[bool, dict]: |
| data: dict = {"items": []} |
| if not answer: |
| return False, data |
|
|
| s = _strip_json_fence(strip_thinking(answer)) |
|
|
| parsed = None |
| try: |
| parsed = json.loads(s) |
| except Exception: |
| parsed = None |
|
|
| if parsed is None: |
| lb, rb = s.find("["), s.rfind("]") |
| if lb != -1 and rb != -1 and rb > lb: |
| try: |
| parsed = json.loads(s[lb : rb + 1]) |
| except Exception: |
| parsed = None |
|
|
| if parsed is None: |
| for m in re.finditer(r"\[[\s\S]*?\]", s): |
| try: |
| cand = json.loads(m.group(0)) |
| if isinstance(cand, list): |
| parsed = cand |
| break |
| except Exception: |
| continue |
|
|
| if parsed is None or not isinstance(parsed, list): |
| return False, data |
|
|
| items: list[dict] = [] |
| valid_all = True |
| for i, it in enumerate(parsed, start=1): |
| if not isinstance(it, dict): |
| valid_all = False |
| continue |
| bbox = it.get("bbox") |
| if not isinstance(bbox, list) or len(bbox) != 4 or not all(isinstance(v, (int, float)) for v in bbox): |
| valid_all = False |
| continue |
| x1, y1, x2, y2 = [int(round(float(v))) for v in bbox] |
| x1 = max(0, min(1000, x1)) |
| y1 = max(0, min(1000, y1)) |
| x2 = max(0, min(1000, x2)) |
| y2 = max(0, min(1000, y2)) |
| if not (x1 < x2 and y1 < y2): |
| valid_all = False |
| continue |
|
|
| char = it.get("char", "") |
| if not isinstance(char, str): |
| char = str(char) if char is not None else "" |
| char = char.strip() |
|
|
| idx = it.get("idx", i) |
| items.append( |
| { |
| "idx": int(idx) if isinstance(idx, (int, float)) else i, |
| "bbox": [x1, y1, x2, y2], |
| "char": char, |
| } |
| ) |
|
|
| data["items"] = items |
| if len(parsed) == 0: |
| return True, data |
| if not items: |
| return False, data |
| return valid_all, data |
|
|