feat: 新增命令行 API 客户端,修复发布配置改写的反斜杠 bug
scripts/api_client.py:完整走一遍 upload -> 轮询 i2t -> 打印识别文本与抽取字段。 只用标准库(不依赖 requests),可直接拷到别的机器上跑;支持多图、 --min-confidence(服务端过滤)、--json、--health。参数校验放在探活之前, 参数写错时不会因为连不上服务而报出误导性错误。 scripts/prepare_release_config.py:re.sub 的**替换串**里反斜杠是转义序列, 写 Windows 路径(model_dir=D:\...)必然抛 "bad escape \c"。改用函数式替换。 .gitignore 补 *.zip:打包产物压出来有数 GB,别误提交。 README:补命令行客户端用法、发布包共用项目 models 目录的做法。 Co-Authored-By: Claude Code <noreply@anthropic.com>
This commit is contained in:
@@ -15,3 +15,6 @@ http/fixtures/*.png
|
||||
dist*/
|
||||
build/
|
||||
*.spec
|
||||
|
||||
# 打包产物的压缩包(体积可达数 GB),不要进仓库
|
||||
*.zip
|
||||
|
||||
@@ -8,22 +8,21 @@
|
||||
ocr-server/
|
||||
├── config.yaml # 项目配置文件(启动前修改)
|
||||
├── http/ocr-server.http # IDEA / IntelliJ HTTP Client 测试请求
|
||||
├── models/ # 可随发布包分发的 PP-OCRv5 模型(约 166 MB)
|
||||
│ └── official_models/
|
||||
│ ├── PP-OCRv5_server_det/
|
||||
│ └── PP-OCRv5_server_rec/
|
||||
├── models/ # 预置的 8 套 PP-OCR 模型(约 398 MB)
|
||||
│ └── official_models/ # 发布包默认也读这里(见「模型路径」)
|
||||
├── data/uploads/ # 上传图片运行时目录
|
||||
├── src/ocr_server/
|
||||
├── scripts/start.ps1 # Windows 源码/发布包启动
|
||||
├── scripts/start.bat # Windows 启动快捷方式
|
||||
├── scripts/start.sh # Ubuntu 源码/发布包启动
|
||||
├── scripts/start.ps1 # Windows 源码/发布包启动
|
||||
├── scripts/start.bat # Windows 启动快捷方式
|
||||
├── scripts/start.sh # Ubuntu 源码/发布包启动
|
||||
├── scripts/package-windows.ps1 # Windows exe 打包
|
||||
├── scripts/package-ubuntu.sh # Ubuntu ELF 打包
|
||||
├── scripts/packaged_entry.py # PyInstaller 入口
|
||||
├── scripts/package-ubuntu.sh # Ubuntu ELF 打包
|
||||
├── scripts/download-models.bat # Windows 下载/补齐模型
|
||||
├── scripts/download-models.sh # Ubuntu/Linux 下载/补齐模型
|
||||
├── scripts/download_models.py # Python 下载入口
|
||||
└── scripts/benchmark_ocr.py # 测量当前配置的 OCR 耗时
|
||||
├── scripts/switch_env.py # 建立/切换 CPU 与 GPU 两套虚拟环境
|
||||
├── scripts/install_paddle.py # 按显卡安装 CPU/GPU 版 paddlepaddle
|
||||
├── scripts/download_models.py # 下载/补齐预置模型
|
||||
├── scripts/benchmark_ocr.py # 测量当前配置的 OCR 耗时与召回
|
||||
└── scripts/api_client.py # 命令行调用 OCR 接口
|
||||
```
|
||||
|
||||
## 模型位置和下载策略
|
||||
@@ -932,6 +931,35 @@ extract_include_empty: true
|
||||
|
||||
抽取发生在查询时(`/api/v1/i2t`),基于 `min_confidence` 过滤后的文本,因此改预设或调 `min_confidence` 不需要重新识别。`extract_fields` 留空(或删除该配置)即关闭此功能。
|
||||
|
||||
### 命令行客户端
|
||||
|
||||
不想记接口细节就用 `scripts/api_client.py`,它只用标准库,不依赖 requests,可直接拷到别的机器上跑:
|
||||
|
||||
```bash
|
||||
# 服务默认在 http://127.0.0.1:8000,可用 --host 指定别的地址
|
||||
python scripts/api_client.py 图片.png
|
||||
python scripts/api_client.py 图片1.png 图片2.png # 多张依次处理
|
||||
python scripts/api_client.py 图片.png --min-confidence 0.6 # 只要置信度 >= 0.6 的片段
|
||||
python scripts/api_client.py 图片.png --json # 打印原始 JSON 响应
|
||||
python scripts/api_client.py --health # 只探活
|
||||
```
|
||||
|
||||
输出示例:
|
||||
|
||||
```text
|
||||
服务 http://127.0.0.1:8000 正常:{'code': 0, 'message': 'success', 'data': {'status': 'ok'}}
|
||||
[general_ocr_002.png]
|
||||
耗时 462.45 ms,34 个文本片段
|
||||
── 识别文本 ────────────────────────────────────────
|
||||
PASS BOARDING 登机牌
|
||||
SEAT NO. 座位号 SERIAL NO. CLASS 序号 舱位
|
||||
...
|
||||
```
|
||||
|
||||
它做的事就是完整走一遍 `POST /api/v1/upload` → 轮询 `POST /api/v1/i2t`,把排队位置和进度打在
|
||||
同一行,结束后打印识别文本与抽取到的字段。`--timeout` 可调单张最长的等待秒数(默认 300,
|
||||
GPU 首次推理有 10~20 秒预热,别设太短)。
|
||||
|
||||
## IDEA HTTP Client 测试
|
||||
|
||||
已提供:
|
||||
@@ -1350,6 +1378,31 @@ uv run python scripts/download_models.py
|
||||
可以从 `src/ocr_server/ocr.py` 的 `STAGED_DET_MODELS` / `STAGED_REC_MODELS`
|
||||
里删掉再重新下载,打包时 `models/` 目录里没有的自然不会被带上。
|
||||
|
||||
#### 让发布包共用项目里的 models 目录
|
||||
|
||||
不想让每个发布包各带一份 398MB 模型,可以让它们都指向同一份。**只改发布包里的
|
||||
`config.yaml` 就行,不用重新打包**——`model_dir` 写绝对路径时原样使用:
|
||||
|
||||
```yaml
|
||||
model_dir: D:\code\work\ocr-server\models
|
||||
```
|
||||
|
||||
改完启动日志里能直接看到用的是哪份模型,便于确认:
|
||||
|
||||
```text
|
||||
INFO [ocr_server] 启动 OCR 服务 ... model_dir=D:\code\work\ocr-server\models
|
||||
```
|
||||
|
||||
当前 `dist/ocr-server/` 与 `dist/ocr-server-gpu/` 已经这样配好了。注意这条路要求
|
||||
**该目录在运行服务的机器上确实存在**;包内原本复制的 `models/` 此时不再被读取,
|
||||
想省 398MB 可以删掉。
|
||||
|
||||
改发布配置用现成的脚本(它会保持 UTF-8,不会写坏中文注释):
|
||||
|
||||
```bash
|
||||
python scripts/prepare_release_config.py dist/ocr-server/config.yaml dist/ocr-server/config.yaml "model_dir=D:\code\work\ocr-server\models"
|
||||
```
|
||||
|
||||
## 测试
|
||||
|
||||
```bash
|
||||
|
||||
@@ -0,0 +1,188 @@
|
||||
"""调用 OCR 服务的命令行客户端:上传图片 -> 轮询结果 -> 打印识别内容。
|
||||
|
||||
只用标准库,不依赖 requests/httpx,方便在任意机器上直接跑。
|
||||
|
||||
用法:
|
||||
python scripts/api_client.py 图片.png
|
||||
python scripts/api_client.py 图片1.png 图片2.png # 多张依次处理
|
||||
python scripts/api_client.py 图片.png --min-confidence 0.6 # 过滤低置信片段
|
||||
python scripts/api_client.py 图片.png --json # 打印原始响应
|
||||
python scripts/api_client.py --health # 只探活
|
||||
python scripts/api_client.py 图片.png --host http://192.168.1.10:8000
|
||||
|
||||
服务端接口(见 README「API」):
|
||||
POST /api/v1/upload 上传图片,返回 file_id,任务进入队列
|
||||
POST /api/v1/i2t 用 file_id 查结果;未完成时返回 queued/processing 与进度
|
||||
GET /healthz 健康检查
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import mimetypes
|
||||
import sys
|
||||
import time
|
||||
import urllib.error
|
||||
import urllib.request
|
||||
import uuid
|
||||
from pathlib import Path
|
||||
|
||||
DEFAULT_HOST = "http://127.0.0.1:8000"
|
||||
POLL_INTERVAL_SECONDS = 0.5
|
||||
FINISHED_STATUSES = ("completed", "failed")
|
||||
|
||||
|
||||
def post_json(url: str, payload: dict, timeout: float) -> dict:
|
||||
request = urllib.request.Request(
|
||||
url,
|
||||
data=json.dumps(payload).encode("utf-8"),
|
||||
headers={"Content-Type": "application/json"},
|
||||
method="POST",
|
||||
)
|
||||
with urllib.request.urlopen(request, timeout=timeout) as response:
|
||||
return json.loads(response.read().decode("utf-8"))
|
||||
|
||||
|
||||
def upload_image(url: str, image_path: Path, timeout: float) -> dict:
|
||||
boundary = "----ocrclient" + uuid.uuid4().hex
|
||||
content_type = mimetypes.guess_type(image_path.name)[0] or "application/octet-stream"
|
||||
head = (
|
||||
f"--{boundary}\r\n"
|
||||
f'Content-Disposition: form-data; name="file"; filename="{image_path.name}"\r\n'
|
||||
f"Content-Type: {content_type}\r\n\r\n"
|
||||
).encode("utf-8")
|
||||
tail = f"\r\n--{boundary}--\r\n".encode("utf-8")
|
||||
body = head + image_path.read_bytes() + tail
|
||||
|
||||
request = urllib.request.Request(
|
||||
url,
|
||||
data=body,
|
||||
headers={"Content-Type": f"multipart/form-data; boundary={boundary}"},
|
||||
method="POST",
|
||||
)
|
||||
with urllib.request.urlopen(request, timeout=timeout) as response:
|
||||
return json.loads(response.read().decode("utf-8"))
|
||||
|
||||
|
||||
def wait_for_result(
|
||||
host: str, file_id: str, timeout: float, quiet: bool, min_confidence: float
|
||||
) -> dict:
|
||||
"""轮询 /api/v1/i2t 直到任务结束。中途把进度打到同一行。"""
|
||||
deadline = time.monotonic() + timeout
|
||||
last_progress = -1
|
||||
while time.monotonic() < deadline:
|
||||
payload = post_json(
|
||||
f"{host}/api/v1/i2t",
|
||||
{"file_id": file_id, "min_confidence": min_confidence},
|
||||
timeout=30,
|
||||
)
|
||||
data = payload.get("data") or {}
|
||||
status = data.get("status")
|
||||
if status in FINISHED_STATUSES:
|
||||
if not quiet and last_progress >= 0:
|
||||
print() # 结束进度行
|
||||
return data
|
||||
if not quiet:
|
||||
progress = data.get("progress", 0)
|
||||
extra = ""
|
||||
if status == "queued" and data.get("queue_position"):
|
||||
extra = f" 排队第 {data['queue_position']} 位"
|
||||
if progress != last_progress:
|
||||
print(f"\r 进度 {progress:>3}% {status}{extra} ", end="", flush=True)
|
||||
last_progress = progress
|
||||
time.sleep(POLL_INTERVAL_SECONDS)
|
||||
raise TimeoutError(f"等待超过 {timeout:.0f} 秒,任务仍未结束")
|
||||
|
||||
|
||||
def print_result(data: dict, show_json: bool) -> None:
|
||||
if show_json:
|
||||
print(json.dumps(data, ensure_ascii=False, indent=2))
|
||||
return
|
||||
|
||||
result = data.get("result") or {}
|
||||
text = result.get("text", "")
|
||||
polygon = result.get("polygon") or {}
|
||||
print(f" 耗时 {data.get('ocr_elapsed_ms')} ms,{len(polygon)} 个文本片段")
|
||||
if text:
|
||||
print(" ── 识别文本 " + "─" * 40)
|
||||
for line in text.splitlines():
|
||||
print(f" {line}")
|
||||
fields = data.get("fields")
|
||||
if fields:
|
||||
print(" ── 抽取字段 " + "─" * 40)
|
||||
for key, value in fields.items():
|
||||
print(f" {key}: {value}")
|
||||
|
||||
|
||||
def process_one(host: str, image_path: Path, args: argparse.Namespace) -> bool:
|
||||
print(f"[{image_path.name}]")
|
||||
if not image_path.is_file():
|
||||
print(f" 文件不存在:{image_path}", file=sys.stderr)
|
||||
return False
|
||||
|
||||
try:
|
||||
uploaded = upload_image(f"{host}/api/v1/upload", image_path, timeout=args.timeout)
|
||||
except urllib.error.HTTPError as exc:
|
||||
detail = exc.read().decode("utf-8", "ignore")
|
||||
print(f" 上传失败 HTTP {exc.code}:{detail}", file=sys.stderr)
|
||||
return False
|
||||
except urllib.error.URLError as exc:
|
||||
print(f" 连不上服务 {host}:{exc.reason}", file=sys.stderr)
|
||||
return False
|
||||
|
||||
file_id = (uploaded.get("data") or {}).get("file_id")
|
||||
if not file_id:
|
||||
print(f" 上传响应异常:{uploaded}", file=sys.stderr)
|
||||
return False
|
||||
|
||||
try:
|
||||
data = wait_for_result(host, file_id, args.timeout, args.quiet, args.min_confidence)
|
||||
except TimeoutError as exc:
|
||||
print(f" {exc}", file=sys.stderr)
|
||||
return False
|
||||
|
||||
if data.get("status") == "failed":
|
||||
print(f" 识别失败:{data.get('error')}", file=sys.stderr)
|
||||
return False
|
||||
print_result(data, args.json)
|
||||
return True
|
||||
|
||||
|
||||
def main() -> int:
|
||||
parser = argparse.ArgumentParser(description="OCR 服务命令行客户端")
|
||||
parser.add_argument("images", nargs="*", type=Path, help="要识别的图片路径")
|
||||
parser.add_argument("--host", default=DEFAULT_HOST, help=f"服务地址,默认 {DEFAULT_HOST}")
|
||||
parser.add_argument("--timeout", type=float, default=300.0, help="单张图片最长等待秒数")
|
||||
parser.add_argument("--json", action="store_true", help="打印完整 JSON 响应")
|
||||
parser.add_argument("--quiet", action="store_true", help="不打印轮询进度")
|
||||
parser.add_argument("--health", action="store_true", help="只做健康检查")
|
||||
parser.add_argument(
|
||||
"--min-confidence",
|
||||
type=float,
|
||||
default=0.0,
|
||||
help="只返回置信度不低于该值的片段(服务端过滤),默认 0 表示全要",
|
||||
)
|
||||
args = parser.parse_args()
|
||||
host = args.host.rstrip("/")
|
||||
|
||||
# 先校验参数再探活:参数写错时不该因为连不上服务而报出误导性的错误。
|
||||
if not args.images and not args.health:
|
||||
parser.error("请至少给一张图片路径(或用 --health 只探活)")
|
||||
|
||||
try:
|
||||
with urllib.request.urlopen(f"{host}/healthz", timeout=10) as response:
|
||||
print(f"服务 {host} 正常:{json.loads(response.read().decode('utf-8'))}")
|
||||
except urllib.error.HTTPError as exc:
|
||||
print(f"服务 {host} 响应异常:HTTP {exc.code}", file=sys.stderr)
|
||||
return 1
|
||||
except urllib.error.URLError as exc:
|
||||
print(f"连不上服务 {host}:{exc.reason}", file=sys.stderr)
|
||||
return 1
|
||||
if args.health:
|
||||
return 0
|
||||
|
||||
failed = sum(0 if process_one(host, path, args) else 1 for path in args.images)
|
||||
return 1 if failed else 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
@@ -17,7 +17,9 @@ from pathlib import Path
|
||||
def apply_override(text: str, key: str, value: str) -> str:
|
||||
pattern = re.compile(rf"^{re.escape(key)}:.*$", re.MULTILINE)
|
||||
if pattern.search(text):
|
||||
return pattern.sub(f"{key}: {value}", text, count=1)
|
||||
# 必须用函数式替换:re.sub 会把替换串里的反斜杠当转义序列,
|
||||
# Windows 路径(D:\code\...)会直接抛 "bad escape \c"。
|
||||
return pattern.sub(lambda _match: f"{key}: {value}", text, count=1)
|
||||
# 源文件里没有这一项(例如被注释掉了),追加到末尾。
|
||||
return text.rstrip("\n") + f"\n{key}: {value}\n"
|
||||
|
||||
|
||||
@@ -9,6 +9,7 @@ PREPARE_CONFIG = PROJECT_ROOT / "scripts" / "prepare_release_config.py"
|
||||
PADDLE_ENV = PROJECT_ROOT / "scripts" / "paddle_env.py"
|
||||
INSTALL_PADDLE = PROJECT_ROOT / "scripts" / "install_paddle.py"
|
||||
SWITCH_ENV = PROJECT_ROOT / "scripts" / "switch_env.py"
|
||||
API_CLIENT = PROJECT_ROOT / "scripts" / "api_client.py"
|
||||
|
||||
SAMPLE = """\
|
||||
# 中文注释必须原样保留
|
||||
@@ -101,6 +102,24 @@ def test_install_paddle_rejects_unknown_cuda_channel() -> None:
|
||||
assert result.returncode == 2
|
||||
|
||||
|
||||
def test_api_client_help_and_missing_args() -> None:
|
||||
assert run_script(API_CLIENT, "--help").returncode == 0
|
||||
# 不给图片又不带 --health:argparse 应该报错退出(2)
|
||||
assert run_script(API_CLIENT).returncode == 2
|
||||
|
||||
|
||||
def test_api_client_fails_cleanly_when_server_unreachable() -> None:
|
||||
"""服务不可达时要干净退出并给出可读提示,而不是抛未捕获的异常。
|
||||
|
||||
不断言具体是「连接被拒」还是「HTTP 错误」:CI/沙箱里可能有透明代理,
|
||||
两者都会出现。真正要保证的是有明确报错、没有 traceback。
|
||||
"""
|
||||
result = run_script(API_CLIENT, "--health", "--host", "http://127.0.0.1:1")
|
||||
assert result.returncode == 1
|
||||
assert "http://127.0.0.1:1" in result.stderr
|
||||
assert "Traceback" not in result.stderr
|
||||
|
||||
|
||||
def test_switch_env_list_reports_both_variants() -> None:
|
||||
"""--list 只读,不能改动任何环境。"""
|
||||
result = run_script(SWITCH_ENV, "--list")
|
||||
|
||||
Reference in New Issue
Block a user