superrob1500
/
mirror-immich
mirror of https://github.com/immich-app/immich
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

align ocr detection normalization with paddle ppocrv5

pull/27774/head
Huan Yu 2026-04-13 22:18:33 +08:00
parent d377d2e145
commit 057c7629bd
1 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
machine-learning/immich_ml/models/ocr/detection.py
View File

@ -25,8 +25,12 @@ class TextDetector(InferenceModel):
def __init__(self, model_name: str, min_score: float = 0.5, **model_kwargs: Any) -> None:
super().__init__(model_name.split("__")[-1], **model_kwargs, model_format=ModelFormat.ONNX)
self.max_resolution = 736
self.mean = np.array([0.5, 0.5, 0.5], dtype=np.float32)
self.std_inv = np.float32(1.0) / (np.array([0.5, 0.5, 0.5], dtype=np.float32) * 255.0)
# Align with Paddle NormalizeImage:
# scale=1/255, mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
# This implementation works on raw 0..255 pixels, so we fold scale into mean/std:
# (x/255 - mean) / std == (x - mean*255) / (std*255)
self.mean = np.array([0.485, 0.456, 0.406], dtype=np.float32) * 255.0
self.std_inv = np.float32(1.0) / (np.array([0.229, 0.224, 0.225], dtype=np.float32) * 255.0)
self._empty: TextDetectionOutput = {
"boxes": np.empty(0, dtype=np.float32),
"scores": np.empty(0, dtype=np.float32),