Skip to content

Latest commit

 

History

History
64 lines (52 loc) · 3.92 KB

README.md

File metadata and controls

64 lines (52 loc) · 3.92 KB
Raw

Custom node for object detection models from OpenVINO Model Zoo

This custom node analyses the response of models from OpenVINO Model Zoo. Based on the inference results and the original image, it generates a list of detected boxes for following object recognition models. Each image in the output will be resized to the predefined target size to fit the input of the next model in the DAG pipeline. Additionally to the detected text boxes, two additional outputs are returned - information about coordinates and confidence levels of each box detection.

Supported models

All OpenVINO Model Zoo object detection models with specific output tensor detection with shape: [1, 1, 200, 7]:

  • face-detection
  • face-detection-adas
  • face-detection-retail
  • person-detection
  • person-detection-adas
  • person-detection-retail
  • vehicle-detection
  • vehicle-detection-adas
  • product-detection
  • person-vehicle-bike-detection
  • vehicle-license-plate-detection
  • pedestrian-and-vehicle-detector

Building custom node library

You can build the shared library of the custom node simply by running command in this custom node folder context:

make

It will compile the library inside a docker container and save the results in lib folder.

Custom node inputs

Input name Description Shape Precision
image Input image in an array format. Only batch size 1 is supported and images must have 3 channels. Resolution is configurable via parameters original_image_width and original_image_height. Color data required only in BGR format. 1,3,H,W FP32
detection object detection model output 1,1,200,7 FP32

Custom node outputs

Output name Description Shape Precision
images Returns images representing detected text boxes. Boxes are filtered based on confidence_threshold param. Resolution is defined by the node parameters. All images are in a single batch. Batch size depend on the number of detected objects. N,1,C,H,W FP32
coordinates For every detected box N the following info is added: x coordinate for the box center, y coordinate for the box center, box original width, box original height N,1,4 I32
confidences For every detected box N information about confidence level (N - number of detected boxes; more about demultiplexing here) N,1,1 FP32

Custom node parameters

Parameters can be defined in pipeline definition in OVMS configuration file. Read more about node parameters.

Parameter Description Default Required
original_image_width Required input image width ✓
original_image_height Required input image height ✓
target_image_width Target width of the boxes in output. Boxes in the original image will be resized to that value. ✓
target_image_height Target width of the boxes in output. Boxes in the original image will be resized to that value. ✓
original_image_layout Defines input image layout NCHW
target_image_layout Defines the data layout of detected object images in the node output NCHW
convert_to_gray_scale Defines if output images should be in grayscale or in color false
confidence_threshold Number in a range of 0-1 ✓
debug Defines if debug messages should be displayed false
max_output_batch Prevents too big batches with incorrect confidence level. It can avoid exceeding RAM resources 100
filter_label_id For object detection models with multiple label IDs results, use this parameter to filter the ones with desired ID