Skip to content

Workflow

The Image Processing and Duplicate Detection workflow is designed to provide reliable face detection, recognition, and duplicate detection by leveraging a pre-trained deep learning model.

Inference Mode Operation

This application operates strictly in inference mode, which means that it does not perform training but instead relies on a pre-trained model for face recognition tasks. This mode ensures that the application can rapidly deploy face recognition capabilities without the computational cost or time required for training models from scratch.

Pre-Trained Model Usage.

The pre-trained model is stored in Azure Blob Storage and is automatically downloaded by the application when it starts. This process ensures that the latest version of the model is always available for inference.

Manual Model Update.

In addition to automatic loading, administrators have the option to manually update the model through the admin panel. This feature provides flexibility for applying updates or new models when improvements or changes are required without modifying the underlying code.

Model Details

The face recognition capabilities are powered by the OpenCV ⧉ library. Currently, the application utilizes an open-source, pre-trained model specifically designed for face detection.

Model Components

  • deploy.prototxt: This file defines the model architecture, including the network layers and the specific parameters used for each layer. It serves as a blueprint that guides how the model processes input data.
  • res10_300x300_ssd_iter_140000.caffemodel: This file contains the trained weights of the model. It was trained using the Caffe deep learning framework, with a total of 140,000 iterations, ensuring robustness in face detection tasks.

Model Architecture

  • The model follows the Res10 architecture, which is known for its efficiency in detecting faces. Res10 is a lightweight model that balances speed and accuracy, making it suitable for real-time applications.
  • The model operates with a fixed input resolution of 300x300, optimizing detection for faces within that scale. This resolution offers a compromise between detail and processing efficiency, allowing the model to quickly identify facial features without excessive computational load.
  • SSD Methodology. The model utilizes the Single Shot MultiBox Detector (SSD) methodology, which is a popular approach for object detection. SSD is designed to predict both the bounding boxes and the confidence scores for each object in a single forward pass through the network. By leveraging the SSD approach, the model can efficiently detect multiple faces in a single image, making it suitable for batch processing and applications where rapid detection is required.

Worklow Diagram

The workflow diagram illustrates the overall process of Image Processing and Duplicate Detection within the system, showcasing how different components interact to achieve face detection, recognition, and duplicate identification. 

flowchart LR
  subgraph DNNManager[DNN Manager]
      direction TB
      load_model[Load Model] -- computation <a href="../config/#dnn_backend">backend</a>\ntarget <a href="../config/#dnn_target">device</a>  --> set_preferences[Set Preferences]
  end

  subgraph ImageProcessing[Image Processing]
      direction LR

      subgraph FaceDetection[Face Detection]
          direction TB
          load_image[Load Image] -- decoded image as 3D numpy array\n(height, width, channels of BlueGreeRed color space) --> prepare_image[Prepare Image] -- blob 4D tensor\n(normalized size, use <a href="../config/#blob_from_image_scale_factor">scale factor</a> and <a href="../config/#blob_from_image_mean_values">means</a>) --> run_model[Run Model] -- shape (1, 1, N, 7),\n1 image\nN is the number of detected faces\neach face is described by the 7 detection values--> filter_results[Filter Results] -- <a href="../config/#face_detection_confidence">confidence</a> is above the minimum threshold,\n<a href="../config/#nms_threshold">NMS</a> to suppress overlapping bounding boxes --> return_detections[Return Detections]
      end

      subgraph FaceRecognition[Face Recognition]
          direction TB
          load_image_[Load Image] --> detect_faces[Detect Faces] -- detected face regions\n<a href="../config/#face_encodings_num_jitters">number of times</a> to re-sample the face\n<a href="../config/#face_encodings_model">key facial landmarks</a> --> generate_encodings[Generate Encodings] -- numerical representations of the facial features\n(face's geometry and appearance) --> save_encodings[Save Encodings]
      end
  end

  subgraph DuplicateFinder[Duplicate Finder]
      direction TB
      load_encodings[Load Encodings] --> compare_encodings[Compare Encodings] -- face distance less then <a href="../config/#face_distance_threshold">threshold</a> --> return_duplicates[Return Duplicates]
  end

  DNNManager --> ImageProcessing --> DuplicateFinder
  FaceDetection --> FaceRecognition
Press "Alt" / "Option" to enable Pan & Zoom