105 lines
3.6 KiB
Python
105 lines
3.6 KiB
Python
"""
|
|
OCR 请求 / 响应 Pydantic 模型。
|
|
|
|
定义天目微服务内部的 API 契约，供 endpoints 和 engine 使用。
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from typing import Any
|
|
|
|
from pydantic import BaseModel, Field
|
|
|
|
|
|
class OCRRequest(BaseModel):
|
|
"""通用 OCR 请求。"""
|
|
|
|
image: str = Field(description="图片数据：base64 编码字符串或 HTTP(s) URL")
|
|
engine: str = Field(
|
|
default="auto",
|
|
description="引擎选择：auto（自动选择）/ ppocrv6 / paddleocr-vl",
|
|
)
|
|
lang: list[str] = Field(
|
|
default_factory=lambda: ["ch", "en"],
|
|
description="识别语言列表，默认中英文。仅对 ppocrv6 引擎生效",
|
|
)
|
|
resize: int | None = Field(
|
|
default=None,
|
|
description="可选：图片最大边长（px），超过则等比缩放。减少传输 / 推理耗时",
|
|
)
|
|
save_md: bool = Field(
|
|
default=False,
|
|
description="是否将识别结果保存为 Markdown 文件到 output/md/ 目录",
|
|
)
|
|
|
|
|
|
class BBox(BaseModel):
|
|
"""文本框坐标。"""
|
|
|
|
points: list[list[float]] = Field(description="四个顶点坐标 [[x1,y1],[x2,y1],[x3,y2],[x4,y2]]")
|
|
|
|
|
|
class TextBlock(BaseModel):
|
|
"""单行/单块识别结果。"""
|
|
|
|
text: str = Field(description="识别文字")
|
|
confidence: float = Field(description="置信度 0~1", ge=0, le=1)
|
|
bbox: BBox = Field(description="文本框坐标")
|
|
|
|
|
|
class OCRResponse(BaseModel):
|
|
"""通用 OCR 响应。"""
|
|
|
|
engine: str = Field(description="实际使用的引擎，如 ppocrv6 / paddleocr-vl")
|
|
text: str = Field(default="", description="拼接后的全文")
|
|
blocks: list[TextBlock] = Field(default_factory=list, description="逐块识别结果")
|
|
duration_ms: float = Field(default=0, description="推理耗时（毫秒）")
|
|
error: str | None = Field(default=None, description="错误信息，成功时为 null")
|
|
saved_md_path: str | None = Field(
|
|
default=None, description="保存的 Markdown 文件路径（仅 save_md=true 时有值）"
|
|
)
|
|
|
|
|
|
class OCRVLRequest(BaseModel):
|
|
"""PaddleOCR-VL 复杂文档解析请求。"""
|
|
|
|
image: str = Field(description="图片数据：base64 编码字符串或 HTTP(s) URL")
|
|
output_format: str = Field(
|
|
default="markdown",
|
|
description="输出格式：markdown / json / text",
|
|
)
|
|
save_md: bool = Field(
|
|
default=False,
|
|
description="是否将识别结果保存为 Markdown 文件到 output/md/ 目录",
|
|
)
|
|
|
|
|
|
class OCRVLResponse(BaseModel):
|
|
"""PaddleOCR-VL 响应。"""
|
|
|
|
engine: str = Field(default="paddleocr-vl")
|
|
text: str = Field(default="", description="结构化输出文本（Markdown / JSON / 纯文本）")
|
|
format: str = Field(default="markdown", description="实际输出格式")
|
|
duration_ms: float = Field(default=0, description="推理耗时（毫秒）")
|
|
error: str | None = Field(default=None, description="错误信息，成功时为 null")
|
|
saved_md_path: str | None = Field(
|
|
default=None, description="保存的 Markdown 文件路径（仅 save_md=true 时有值）"
|
|
)
|
|
|
|
|
|
# ── 健康检查 ──────────────────────────────────────────────
|
|
|
|
|
|
class EngineStatus(BaseModel):
|
|
"""引擎状态。"""
|
|
|
|
status: str = Field(description="ok / degraded / error / unavailable")
|
|
|
|
|
|
class HealthResponse(BaseModel):
|
|
"""健康检查响应。"""
|
|
|
|
status: str = Field(description="整体状态：ok / degraded / error")
|
|
engines: dict[str, Any] = Field(default_factory=dict, description="各引擎状态")
|
|
version: str = Field(description="天目版本")
|