returns a func to create a person_data dict for testing

Preds and prints the acc of a sklearn clsf

Load file from fpath and return as bytes

Returns the file extension for the given mode. Args: mode (str): Either "image" or "video". Returns: str: The file extens

Note: only one image is assumed for inference Returns a dict with params status and message along with optionally other parameters status

Gets the milvus collection connection with the given collection name otherwise creates a new one Note: index_metric_params: dict = {"nlist":

Insert person_data into mysql table with param binding Note: the transaction must be committed after if commit is False

run func and get results as dict

Caches the given data to a local file. Args: file_cache_path (str): The path to save the data. data (bytes): The data to cach

Delete record from mysql db using the uniq person_id

Downloads the file at the given URL to the local path using requests library. This version uses streaming to handle large files and includes

Encrypts & returns vector obj

Get registered person by person_id. Checks redis cache, otherwise query mysql

Calculates the encrypted squared l2 norm between encrypted vectors vec_1 and vec_2

Creates a mock for download_url_file that uses local file content

run func and get results as dict

Query mysql db to get all person data

Query mysql db to get full person data using the uniq person_id

Shuffle two numpy arrays in unison

Clip bounding xyxy bounding boxes to image shape (height, width)

Connects to client HE public context. client_public_context is a dict with attributes {s_context, s_public_key, s_relin_key, s_rotate

Draw bboxes on cv2 image boxes must be 2D list/np array of coords xmin, ymin, xmax, ymax foreach bbox confs must be 2D list of co

Extracts input data from media files for inference. Args: FLAGS: The configuration parameters. preprocess_func: The preproces

function that yields face_vectors as an iterator from an img_dir

Get a unique int uuid

Computes the squared l2-dist of vector repr by s_ctx to all vectors in existing vector db And returns a list of bytes of the encrypte

Gets FastAPI application object with CORS enabled.

Creates a Triton Inference Server client and retrieves model metadata and configuration. Args: FLAGS: The configuration parameters.

Get the embedding vector from a facenet model output

Performs inference using the Triton Inference Server and returns the responses. Args: image_data_list: The list of input data.

Get the k nearest neighbors from the dist list i.e. take the min-k dist values and get the respective vectors in vector db

Returns a Pyfhel encryption/decryption object

Get registered data record by person_id. Arguments: person_id: int = number, output_fields: list[str] = ["person_id", "embedd

Initialize Redis/MySQL/Milvus connections with retry/backoff. Safe to call multiple times.

Inserts data in bulk into mysql Each data should match the data model in PersonModel app/models/model.py

Demo inference run

Main function, the img dir containing face images should be passed into the insert_data functions

resize and pad image with color if necessary, maintaining orig scale args: cv2_img: numpy.ndarray = cv2 image new_size: tuple

Parses the model metadata and configuration to extract necessary information for inference. Args: model_metadata: The model metadata.

Postprocess detections & face features from triton-server Args: results: triton server inference result orig_img_size: (width

Register encrypted vector into database. Server must be connected to client first

Generates inference requests for the Triton Inference Server. Args: input_data_list: The list of input data. input_name_list:

Rescale coords (xyxy) from img1_shape to img0_shape img1_shape: (h,w) model input shape img0_shape: (h,w) orig image shape

Train faces from train and test directories with real & fake face images

Use a DNN with pytorch lightining to classify face vectors

Train and eval svm, random forests & adaboost classifiers

Init model feature extractor backbone and final NN classifier

Adds an X-Process-Time header to the response indicating the api request processing time.

Favicon endpoint. Returns the favicon.

Get all registered persons query mysql

Gets all registered persons.

Gets the registered person with the given ID.

Health check endpoint to verify server is running.

Root endpoint. Returns a welcome message.

reference nb: https://github.com/towhee-io/examples/blob/main/image/reverse_image_search/1_build_image_search_engine.ipynb

Inserts embeddings data in bulk into milvus

Yield image paths

load and return an image with a single face

load and return an image with a single face

returns an image url with a single face

load and return an image with a single face

returns an image url with a single face

load and return an image with a single face

Plot one bounding box on image img args bboxes: bounding boxes in xyxy format (x1,y1,x2,y2) img: image in (H,W,C) numpy.ndarr

Preprocess input cv2 image Args: img: cv2 input image width: model input width height: model input height new

Detects faces in image from the file_path and finds the most similar face vector from a set of saved face vectors

recognize person from the face image file uploaded as a file

recognize person from the face image file provided as a url

Detects faces in image from the file_path and saves the face feature vector & the related person_data dict. person_data dict should be ba

registers person face and info with the face uploaded as an image file Person data id: int = must be a unique id in the database, re

registers person face and info with the face image file provided as a url Person data id: int = must be a unique id in the database,

Removes the file at the given path. Args: path (str): The path of the file to remove.

Resizes an image while maintaining aspect ratio. If width and height are both None, no resize is done. If either width or height is None,

If width and height are both None, no resize is done If either width or height is None, resize maintaining aspect

Sets up the async server for httpx>=20, follow_redirects=True (cf. https://github.com/encode/httpx/releases/tag/0.20.0)

Deletes the test person from Milvus. Should call milvus_collec_conn.flush() and milvus_collec_conn.compact(collec_name) after deletion

Deletes test person data from MySQL database.