Skip to content

Latest commit

 

History

History
56 lines (44 loc) · 4.32 KB

README.md

File metadata and controls

56 lines (44 loc) · 4.32 KB
Raw

Safety Gear Detection

Use an optimized and pre-trained MobileNet-SSD neural network to detect people and their safety gear from video input.

How It Works

The sample converts a mobilenet-ssd caffe model for optimized inference and feeds a video frame-by-frame to the OpenVINO Inference Engine. The identified results i.e. detected people and their safety gear (e.g. vest, hardhat) are stored to a text file which are used later to annotate all the frames of the original video.

  • openvino_cgvh_dev_2021.4.dockerfile: Utilizes openvino/ubuntu18_data_dev as the base image and defines configurable runtime environment variables.
  • run_safety_gear_detection.sh: Serves as an entrypoint for the container sample, utilizes the pre-installed model optimizer to convert the mobilenet-ssd caffe model before running inference and annotation python scripts.
  • safety_gear_detection.py: Demonstrates asynchronous inference pipeline on input video file, and saves an output.txt file during execution with resulting bounding box coordinates, detected labels corresponding to IDs from labels.txt, detection probabilities along with perfomance.txt capturing latency and throughput metrics.
  • safety_gear_detection_annotate.py: Reads the original video file, annotates frame-by-frame inference results (e.g. bounding boxes, label text) and saves a new output.mp4 file after execution.

Runtime Configurations

Default Config Description
-e DEVICE=CPU Supports GPU for running on capable integrated GPU.
-e PRECISION=FP16 Will support FP32 model precision in upcoming releases.
-e INPUT_FILE="resources/Safety_Full_Hat_and_Vest.mp4" Input video file path inside the container
-e RUN_ON_PREM="/mount_folder" Directory to save results to e.g. mount point to retrieve logs, results

Build and run on DevCloud

Using the terminal from the DevCloud Coding Environment, navigate to {repo-root}/container-workloads/openvino-dev-latest directory and build:

buildah bud --format docker -f ./developer-samples/python/safety-gear-detection/dockerfile/ubuntu18/openvino_cgvh_dev_2021.4.dockerfile -t $REGISTRY_URL/safety-gear-detection:custom .

Push the container to your devcloud private registry:

buildah push $REGISTRY_URL/safety-gear-detection:custom

Navigate to My Library > Resources and associate the safety-gear-detection:custom resource with a project, configure the Mount Point with /mount_folder and launch.

NOTE:

  • The container playground will ensure GPU access is enabled by default when launching on a device with an integrated GPU.
  • You can indicate the sample to run inference on GPU by passing -e DEVICE=GPU in Configuration Parameters
  • Refer to the developer-guide, configure-imported-containers and select-hardware-and-launch for more information.

Build and run on local system

Navigate to {repo-root}/container-workloads/openvino-dev-latest directory and build:

docker build -f ./developer-samples/python/safety-gear-detection/dockerfile/ubuntu18/openvino_cgvh_dev_2021.4.dockerfile -t safety-gear-detection:custom .

Run the container locally by mounting a local directory to retrieve the results:

docker run --rm -it -e RUN_ON_PREM=/mount_folder -v {PATH-TO-HOST-DIR}:/mount_folder safety-gear-detection:custom

NOTE:

  • To enable GPU access, use runtime sample config by passing -e DEVICE=GPU
  • You must also mount your integrated GPU device e.g. --device /dev/dri:/dev/dri, see openvino/ubuntu18_data_dev for more info.

See README for more info on all marketplace sample applications.