Object Detection

The algorithm uses YOLACT [1] to detect objects in images. The model is assumed to be traced using ImFusion scripts.

1- Bolya, Daniel, Chong Zhou, Fanyi Xiao, and Yong Jae Lee. “Yolact: Real-time instance segmentation.” In Proceedings of the IEEE/CVF international conference on computer vision, pp. 9157-9166. 2019.

Input

A 2D RGB image set or an RGB video stream.

Output

The input images are annotated with detection bounding boxes, classes, instance number and detection scores. Optionally label images containing the class id and instance id of the detected objects are exported.

Description

The following configuration options are possible:

  • Path to model config file: ImFusion machine learning model config file path

  • Non-maximum suppression threshold: Non-maximum suppression threshold

  • Confidence threshold: Confidence threshold for mask priors

  • Score threshold: Score threshold for final detections

  • Max instances: Maximum number of instances to consider

  • Max instances per class: Maximum number of instances per class to consider

  • Class names: Class names, separated by spaces; assumed to be in the same order as the class ids

  • Input size: Size of the input image; this value must be equal to the value in the ImFusion machine learning model config file

  • Export label images: If checked, the class and instance label images will be exported. Currently not supported for video streams.