Face detection

In principle, any self-trained face detection model can be used with DepthAi installed. These models must be converted to a MyriadX blob file format so that they are optimized for best inference on the MyriadX VPU processor and are compatible with DepthAi platform used by the OAK-D-Lite camera. The conversion is done in two steps:

First, the model must be converted from the original framework format to OpenVINO's Intermediate Representation (IR) standard format. To convert models from certain frameworks, you may also need to install additional dependencies, you can find here: https://docs.openvino.ai/2023.1/omz_tools_downloader.html#doxid-omz-tools-downloader. The IR format consists of a .bin file and an .xml file. The .bin file describes the network topology, while the .xml file contains the weights and biases of the model. For this step, OpenVINO's Model Optimizer is used, which can be found here Conversion

Before performing the second step, the color order (RGB or BGR) of the model must be observed and adjusted if necessary, depending on the color order of the training images. The next step is to convert the model from IR format to .blob file format. Either OpenVINO's compile tool or their online blobconverter can be used for this. These approaches can both be found on the same web page as the Model Optimizer.

OpenVINO offers several pre-trained models optimized for the OAK-D-Lite camera. There are models pre-trained by Intel or by the community.

The community trained face detection models found in OpenVINO are: faceboxes-pytorch, face-detection-retail-0044, mobilefacedet-v1-mxnet, mtcnn, retinaface-resnet50-pytorch, mobilenet-ssd, ultra-lightweight-face-detection-rfb-320 and ultra-lightweight-face-detection-slim-320. These models need to be converted to .blob file format beforehand for use on the OAK-D-Lite camera. All models pre-trained by the community can be found here: https://docs.openvino.ai/2023.1/omz_models_group_public.html

The Intel pre-trained face detection models on OpenVINO are: face-detection-adas-0001, face-detection-retail-0004, face-detection-retail-0005, face-detection-0200, face-detection-0202, face-detection-0204, face-detection-0205, face-detection-0206. These models can be downloaded with the OpenVINO model downloader and converted to .blob files with the model converter. These programs can be found here: https://docs.openvino.ai/2023.1/omz_tools_downloader.html#doxid-omz-tools-downloader
All Intel pre-trained models can be found here: https://docs.openvino.ai/2023.1/omz_models_group_intel.html