Smart Classroom C++ G-API Demo

example

The demo shows an example of joint usage of several neural networks to detect student actions (sitting, standing, raising hand for the person-detection-action-recognition-0005 model and sitting, writing, raising hand, standing, turned around, lie on the desk for the person-detection-action-recognition-0006 model) and recognize people by faces in the classroom environment. The demo uses Async API for action and face detection networks. It allows to parallelize execution of face recognition and detection: while face recognition is running on one accelerator, face and action detection could be performed on another. You can use a set of the following pre-trained models with the demo:

  • face-detection-adas-0001, which is a primary detection network for finding faces.

  • landmarks-regression-retail-0009, which is executed on top of the results from the first network and outputs a vector of facial landmarks for each detected face.

  • person-detection-action-recognition-0005, which is a detection network for finding persons and simultaneously predicting their current actions (3 actions - sitting, standing, raising hand).

  • person-detection-action-recognition-0006, which is a detection network for finding persons and simultaneously predicting their current actions (6 actions: sitting, writing, raising hand, standing, turned around, lie on the desk).

  • person-detection-raisinghand-recognition-0001, which is a detection network for finding students and simultaneously predicting their current actions (in contrast with the previous model, predicts only if a student raising hand or not).

  • person-detection-action-recognition-teacher-0002, which is a detection network for finding persons and simultaneously predicting their current actions.

How It Works

On the start-up, the application reads command line parameters and loads four networks for execution on different devices depending on -m... options family. Demo contains two parts. First part is processing faces from gallery. Second part is main detection and recognition functionality. Both parts are graphs. Face gallery graph gets a frame performs inference of Face Detection network. ROI obtained by Face Detector are fed to the Facial Landmarks Regression network. Then landmarks are used to align faces by affine transform and feed them to the Face Recognition network. Output contains identities metrics for each face from gallery. Main graph works with frames from input source. First steps are equal to gallery processing. These are the same operations with Face Detection, Landmarks Regression, Face Recognition networks. Besides this, graph performs Action Detection network. The recognized faces are matched with detected actions to find an action for a recognized person for each frame. Then it draws results on frame. Output contains processed frame and logs.

NOTE: By default, Open Model Zoo demos expect input with BGR channels order. If you trained your model to work with RGB order, you need to manually rearrange the default channels order in the demo application or reconvert your model using the Model Optimizer tool with the --reverse_input_channels argument specified. For more information about the argument, refer to When to Reverse Input Channels section of [Embedding Preprocessing Computation](@ref openvino_docs_MO_DG_Additional_Optimization_Use_Cases).

Preparing to Run

For demo input image or video files you may refer to the Media Files Available for Demos section in the Open Model Zoo Demos overview. The list of models supported by the demo is in <omz_dir>/demos/smart_classroom_demo/cpp_gapi/models.lst file. This file can be used as a parameter for Model Downloader and Converter to download and, if necessary, convert models to OpenVINO IR format (*.xml + *.bin).

An example of using the Model Downloader:

omz_downloader --list models.lst

An example of using the Model Converter:

omz_converter --list models.lst

Supported Models

  • face-detection-adas-0001

  • face-recognition-resnet100-arcface-onnx

  • face-reidentification-retail-0095

  • facenet-20180408-102900

  • landmarks-regression-retail-0009

  • person-detection-action-recognition-0005

  • person-detection-action-recognition-0006

  • person-detection-action-recognition-teacher-0002

  • person-detection-raisinghand-recognition-0001

  • Sphereface

NOTE: Refer to the tables Intel’s Pre-Trained Models Device Support and Public Pre-Trained Models Device Support for the details on models inference support at different devices.

Running

Running the application with the -h option yields the following usage message:

smart_classroom_demo_gapi [OPTION]
Options:

    -h                                          Print a usage message.
    -i                                          Required. An input to process. The input must be a single image, a folder of images, video file or camera id.
    -loop                                       Optional. Enable reading the input in a loop.
    -read_limit                                 Optional. Read length limit before stopping or restarting reading the input.
    -o "<path>"                                 Optional. Name of the output file(s) to save. Frames of odd width or height can be truncated. See https://github.com/opencv/opencv/pull/24086
    -limit "<num>"                              Optional. Number of frames to store in output. If 0 is set, all frames are stored.
    -m_act '<path>'                             Required. Path to the Person/Action Detection Retail model (.xml) file.
    -m_fd '<path>'                              Required. Path to the Face Detection model (.xml) file.
    -m_lm '<path>'                              Required. Path to the Facial Landmarks Regression Retail model (.xml) file.
    -m_reid '<path>'                            Required. Path to the Face Reidentification Retail model (.xml) file.
          Or
    -d_act '<device>'                           Optional. Specify the target device for Person/Action Detection Retail (the list of available devices is shown below). Default value is CPU. The application looks for a suitable plugin for the specified device.
    -d_fd '<device>'                            Optional. Specify the target device for Face Detection Retail (the list of available devices is shown below). Default value is CPU. The application looks for a suitable plugin for the specified device.
    -d_lm '<device>'                            Optional. Specify the target device for Landmarks Regression Retail (the list of available devices is shown below). Default value is CPU. The application looks for a suitable plugin for the specified device.
    -d_reid '<device>'                          Optional. Specify the target device for Face Reidentification Retail (the list of available devices is shown below). Default value is CPU. The application looks for a suitable plugin for the specified device.
    -greedy_reid_matching                       Optional. Use faster greedy matching algorithm in face reid.
    -r                                          Optional. Output Inference results as raw values.
    -ad                                         Optional. Output file name to save per-person action statistics in. Requires -teacher_id and -a_top to be unset and -fg to be set
    -t_ad                                       Optional. Probability threshold for person/action detection.
    -t_ar                                       Optional. Probability threshold for action recognition.
    -t_fd                                       Optional. Probability threshold for face detections.
    -inh_fd                                     Optional. Input image height for face detector.
    -inw_fd                                     Optional. Input image width for face detector.
    -exp_r_fd                                   Optional. Expand ratio for bbox before face recognition.
    -t_reid                                     Optional. Cosine distance threshold between two vectors for face reidentification.
    -fg                                         Optional. Path to a faces gallery in .json format.
    -teacher_id                                 Optional. ID of a teacher. You must also set a faces gallery parameter (-fg) to use it.
    -no_show                                    Optional. Don't show output.
    -min_ad                                     Optional. Minimum action duration in seconds.
    -d_ad                                       Optional. Maximum time difference between actions in seconds.
    -student_ac                                 Optional. List of student actions separated by a comma.
    -top_ac                                     Optional. List of student actions (for top-k mode) separated by a comma.
    -teacher_ac                                 Optional. List of teacher actions separated by a comma.
    -top_id                                     Optional. Target action name.
    -a_top                                      Optional. Number of first K students. If this parameter is positive, the demo detects first K persons with the action, pointed by the parameter 'top_id'
    -crop_gallery                               Optional. Crop images during faces gallery creation.
    -t_reg_fd                                   Optional. Probability threshold for face detections during database registration.
    -min_size_fr                                Optional. Minimum input size for faces during database registration.
    -al                                         Optional. Output file name to save per-person action detections in.
    -ss_t                                       Optional. Number of frames to smooth actions.
    -u                                          Optional. List of monitors to show initially.

Running the application with the empty list of options yields an error message.

Example of a valid command line to run the application with pre-trained models for recognizing students actions:

./smart_classroom_demo_gapi \
    -i <path_to_video> \
    -m_act <path_to_model>/person-detection-action-recognition-0005.xml \
    -student_ac "sitting, standing, raising hand" \
    -m_fd <path_to_model>/face-detection-adas-0001.xml \
    -m_reid <path_to_model>/face-reidentification-retail-0095.xml \
    -m_lm <path_to_model>/landmarks-regression-retail-0009.xml \
    -t_reid 0.8 \
    -fg <path_to_faces_gallery.json>

NOTE: To recognize actions of students, use person-detection-action-recognition-0005 model for 3 basic actions and person-detection-action-recognition-0006 model for 6 actions.

Example of a valid command line to run the application for recognizing actions of a teacher:

./smart_classroom_demo_gapi \
    -i <path_to_video> \
    -m_act <path_to_model>/person-detection-action-recognition-teacher-0002.xml \
    -m_fd <path_to_model>/face-detection-adas-0001.xml \
    -m_reid <path_to_model>/face-reidentification-retail-0095.xml \
    -m_lm <path_to_model>/landmarks-regression-retail-0009.xml \
    -fg <path to faces_gallery.json> \
    -top_id \
    -teacher_id <ID of a teacher in the face gallery>

NOTE: To recognize actions of a teacher, use person-detection-action-recognition-teacher-0002 model. See model description for more details on the list of recognized actions.

Example of a valid command line to run the application for recognizing first raised-hand students:

./smart_classroom_demo_gapi \
  -i <path_to_video> \
  -m_act <path_to_model>/person-detection-raisinghand-recognition-0001.xml \
  -a_top <number of first raised-hand students> \

NOTE: To recognize raising hand action of students, use person-detection-raisinghand-recognition-0001 model. See model description for more details on the list of recognized actions.

Demo Output

The demo uses OpenCV to display the resulting frame with labeled actions and faces. The demo reports:

  • FPS: average rate of video frame processing (frames per second).

You can use these metrics to measure application-level performance.