tensorflow-facedetection

Tensorflow Face Detection

Description

Face detector based on SqueezeNet light (half-channels) as a backbone with a single SSD for indoor/outdoor scenes shot by a front-facing camera. The backbone consists of fire modules to reduce the number of computations. The single SSD head from 1/16 scale feature map has nine clustered prior boxes.

Test

Run OVMS with the model inside

Execute the following command from the pipelines folder:

podman run -d --rm -v ${PWD}/models:/model:Z -p 9000:9000 quay.io/opendatahub/openvino_model_server:stable --model_name face-detection --model_path /model/tensorflow-facedetection --port 9000 --shape auto

Test call

git clone https://github.com/openvinotoolkit/model_server.git
cd model_server/demos/face_detection/python
# Patch the requirements file, because tensorflow 2.11.0 is not working
sed -i 's/tensorflow-serving-api==2.11.0/tensorflow-serving-api==2.13.0/' ../../common/python/requirements.txt
python -m venv .venv
source .venv/bin/activate
pip install -r ../../common/python/requirements.txt
# In case of errors remove the tensorflow-serving-api version from the ../../common/python/requirements.txt
mkdir results
python face_detection.py --batch_size 1 --width 300 --height 300 --grpc_port 9000
# Open the results folder

Credits

• https://docs.openvino.ai/archive/2022.1/omz_models_model_face_detection_retail_0004.html
• https://docs.openvino.ai/2023.2/ovms_demo_face_detection.html