π Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.
# -*- encoding: utf-8 -*-
from pathlib import Path
import pytest
from omegaconf import OmegaConf
from rapidocr.ch_ppocr_rec import TextRecognizer
from rapidocr.ch_ppocr_rec import main as rec_main
from rapidocr.utils.typings import EngineType, ModelType, OCRVersion, TaskType
DICT_URL = "https://example.com/models/ppocrv5_dict.txt"
class StubSession:
"""Session without an embedded character list, like the torch/paddle engines."""
def have_key(self, key: str = "character") -> bool:
return False
@classmethod
def get_dict_key_url(cls, file_info) -> str:
return DICT_URL
def make_rec_cfg(model_root_dir):
return OmegaConf.create(
{
"engine_type": EngineType.TORCH,
"ocr_version": OCRVersion.PPOCRV5,
"task_type": TaskType.REC,
"lang_type": "ch",
"model_type": ModelType.MOBILE,
"rec_keys_path": None,
"model_root_dir": model_root_dir,
}
)
@pytest.fixture
def downloads(monkeypatch):
saved = []
monkeypatch.setattr(
rec_main.DownloadFile, "run", lambda params: saved.append(params.save_path)
)
return saved
def get_character_dict(cfg):
recognizer = TextRecognizer.__new__(TextRecognizer)
recognizer.session = StubSession()
return recognizer.get_character_dict(cfg)
@pytest.mark.parametrize("as_type", [Path, str])
def test_rec_dict_is_downloaded_into_model_root_dir(tmp_path, downloads, as_type):
_, dict_path = get_character_dict(make_rec_cfg(as_type(tmp_path)))
expected = tmp_path / "ppocrv5_dict.txt"
assert Path(dict_path) == expected
assert downloads == [expected]
def test_rec_dict_falls_back_to_package_models_dir(downloads):
_, dict_path = get_character_dict(make_rec_cfg(None))
assert Path(dict_path) == rec_main.DEFAULT_MODEL_PATH / "ppocrv5_dict.txt"
def test_torch_model_path_accepts_str_model_root_dir(tmp_path, monkeypatch):
pytest.importorskip("torch")
from rapidocr.inference_engine.pytorch.networks import main as torch_networks
saved = []
def fake_download(params):
saved.append(params.save_path)
Path(params.save_path).touch()
monkeypatch.setattr(torch_networks.DownloadFile, "run", fake_download)
monkeypatch.setattr(
torch_networks.InferSession,
"get_model_url",
classmethod(
lambda cls, file_info: {
"model_dir": "https://example.com/models/det.pth",
"SHA256": None,
}
),
)
cfg = OmegaConf.create(
{
"model_path": None,
"model_root_dir": str(tmp_path),
"engine_type": EngineType.TORCH,
"ocr_version": OCRVersion.PPOCRV5,
"task_type": TaskType.DET,
"lang_type": "ch",
"model_type": ModelType.MOBILE,
}
)
model_path = torch_networks.ModelLoader._init_model_path(None, cfg)
assert Path(model_path) == tmp_path / "det.pth"
assert saved == [tmp_path / "det.pth"]