Model architecture information¶

The Smart-NMS: user-config file (model architecture) parameters are:

• modelname: modelname is used to choose the decode the subroutine required for smart-nms. String options (“resnet34ssd”, “mv1ssd”, “retinanet”, “yolov3”, “yolov4”, “yolov5”, “resnet18ssd”, “retinaface”)

  Required

  Type: String

• layout: Specifies the layout of output files. String options (“NAWHC”, “NCHW”, “NHWC”, “NCD”, “NDC”)

  Required

  Type: String

• num-classes: Denotes the number of classes in the object-detection model.

  Required

  Type: Int

• num-landmark: Denotes the number of landmark datapoints present in each branch of landmark output given in the landmark-output-list.

  Optional. Default: 0

  Type: Int

• bbox-output-list: List of names of output nodes that contain box features.

  Required

  Type: List

• scores-output-list: List of names of output nodes that contain score/logits features.

  Required

  Type: List

• landmark-output-list: Required if num-landmark > 0, otherwise ignored. List of names of output nodes that contain score/logits features.

  Optional. Default: []

  Type: List

• do-class-specific-nms: Specify True if class-specific NMS is required.

  Optional. Default: False

  Type: Boolean

• do-softmax: Specify True if softmax needs to be performed on outputs from scores-output-list.

  Optional. Default: False

  Type: Boolean

• background-class-idx: Specify index of class which contains background.

  Optional. Default: 0

  Type: Int

• map-classes-coco-81-to-91: Specify True if required to convert class IDs from coco 81 class dataset to coco 91 class dataset.

  Optional. Default: False

  Type: Boolean

• profiling-per-iter: This field enables the profiling of time and memory for each time client call anchor box processing handle. Otherwise, the summary of all the runs will be profiled.

  Optional. Default: False

  Type: Boolean