refactor: migrate person repository to kysely (#15242)

* refactor: migrate person repository to kysely

* `asVector` begone

* linting

* fix metadata faces

* update test

---------

Co-authored-by: Alex <alex.tran1502@gmail.com>
Co-authored-by: mertalev <101130780+mertalev@users.noreply.github.com>

machine-learning/app/models/clip/textual.py

@@ -10,7 +10,7 @@ from tokenizers import Encoding, Tokenizer
 
 from app.config import log
 from app.models.base import InferenceModel
-from app.models.transforms import clean_text
+from app.models.transforms import clean_text, serialize_np_array
 from app.schemas import ModelSession, ModelTask, ModelType
 
 
@@ -18,9 +18,9 @@ class BaseCLIPTextualEncoder(InferenceModel):
     depends = []
     identity = (ModelType.TEXTUAL, ModelTask.SEARCH)
 
-    def _predict(self, inputs: str, **kwargs: Any) -> NDArray[np.float32]:
+    def _predict(self, inputs: str, **kwargs: Any) -> str:
         res: NDArray[np.float32] = self.session.run(None, self.tokenize(inputs))[0][0]
-        return res
+        return serialize_np_array(res)
 
     def _load(self) -> ModelSession:
         session = super()._load()

machine-learning/app/models/clip/visual.py

@@ -10,7 +10,15 @@ from PIL import Image
 
 from app.config import log
 from app.models.base import InferenceModel
-from app.models.transforms import crop_pil, decode_pil, get_pil_resampling, normalize, resize_pil, to_numpy
+from app.models.transforms import (
+    crop_pil,
+    decode_pil,
+    get_pil_resampling,
+    normalize,
+    resize_pil,
+    serialize_np_array,
+    to_numpy,
+)
 from app.schemas import ModelSession, ModelTask, ModelType
 
 
@@ -18,10 +26,10 @@ class BaseCLIPVisualEncoder(InferenceModel):
     depends = []
     identity = (ModelType.VISUAL, ModelTask.SEARCH)
 
-    def _predict(self, inputs: Image.Image | bytes, **kwargs: Any) -> NDArray[np.float32]:
+    def _predict(self, inputs: Image.Image | bytes, **kwargs: Any) -> str:
         image = decode_pil(inputs)
         res: NDArray[np.float32] = self.session.run(None, self.transform(image))[0][0]
-        return res
+        return serialize_np_array(res)
 
     @abstractmethod
     def transform(self, image: Image.Image) -> dict[str, NDArray[np.float32]]:

machine-learning/app/models/facial_recognition/recognition.py

@@ -12,7 +12,7 @@ from PIL import Image
 
 from app.config import log, settings
 from app.models.base import InferenceModel
-from app.models.transforms import decode_cv2
+from app.models.transforms import decode_cv2, serialize_np_array
 from app.schemas import FaceDetectionOutput, FacialRecognitionOutput, ModelFormat, ModelSession, ModelTask, ModelType
 
 
@@ -61,7 +61,7 @@ class FaceRecognizer(InferenceModel):
         return [
             {
                 "boundingBox": {"x1": x1, "y1": y1, "x2": x2, "y2": y2},
-                "embedding": embedding,
+                "embedding": serialize_np_array(embedding),
                 "score": score,
             }
             for (x1, y1, x2, y2), embedding, score in zip(faces["boxes"], embeddings, faces["scores"])

machine-learning/app/models/transforms.py

@@ -4,6 +4,7 @@ from typing import IO
 
 import cv2
 import numpy as np
+import orjson
 from numpy.typing import NDArray
 from PIL import Image
 
@@ -69,3 +70,9 @@ def clean_text(text: str, canonicalize: bool = False) -> str:
     if canonicalize:
         text = text.translate(_PUNCTUATION_TRANS).lower()
     return text
+
+
+# this allows the client to use the array as a string without deserializing only to serialize back to a string
+# TODO: use this in a less invasive way
+def serialize_np_array(arr: NDArray[np.float32]) -> str:
+    return orjson.dumps(arr, option=orjson.OPT_SERIALIZE_NUMPY).decode()