Upload scripts/onnx_ocr.py with huggingface_hub

scripts/onnx_ocr.py

@@ -1,3 +1,47 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:69caefe0a63e8c9fdb2f4c9f635fbee3912b2ed0cac1ab23c983593712cb953c
-size 1309
+#!/usr/bin/env python3
+"""用训练好的位置保留模型识别验证码。输出格式兼容 harness。
+预处理: 灰度->开运算去噪->RGB 160x64 /255 -> (1,3,64,160) -> (1,4,10) -> 4位数字
+"""
+import json
+import os
+import sys
+
+import cv2
+import numpy as np
+import onnxruntime as ort
+from PIL import Image
+
+MODEL = os.path.expanduser("~/sp_captcha_assets/models/captcha_1000.onnx")
+_sess = None
+
+
+def _get_sess():
+    global _sess
+    if _sess is None:
+        _sess = ort.InferenceSession(MODEL, providers=["CPUExecutionProvider"])
+    return _sess
+
+
+def predict(path):
+    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
+    k = np.ones((2, 2), np.uint8)
+    denoised = cv2.morphologyEx(img, cv2.MORPH_OPEN, k)
+    im = Image.fromarray(denoised).convert("RGB").resize((160, 64), Image.BILINEAR)
+    a = np.asarray(im, dtype=np.float32) / 255.0
+    x = a.transpose(2, 0, 1)[None]  # (1,3,64,160)
+    out = _get_sess().run(None, {"input": x})[0]  # (1,4,10)
+    return "".join(str(int(out[0, p].argmax())) for p in range(4))
+
+
+def main():
+    out = {}
+    for f in sys.argv[1:]:
+        try:
+            out[f] = {"onnx": predict(f)}
+        except Exception as e:
+            out[f] = {"onnx": "", "err": str(e)}
+    print(json.dumps(out, ensure_ascii=False))
+
+
+if __name__ == "__main__":
+    main()