Skip to content

decode_yolov3

YOLOv3 decoder.

This layer decodes YOLOv3 into bounding boxes

Parameter

  • img_width : Integer, image width (The default is 416).
  • img_height : Integer, image height (The default is 416).
  • keep_top_k : Integer, maximum num of detected boxes (The default is 300).
  • num_anchors_per_out : Integer, number of anchors per cell (The default is 3).
  • do_nms : Boolean, whether run or not to run NMS (The default is true).
  • nms_thresh : Float, threshold of jaccard overlap for NMS (The default is 0.5).
  • conf_thresh : Float, minimum confidence score to be detected (The default is 0.3).
  • anchors : Float array, sizes of anchors [w1,h1, w2,h2, ...] (default is [10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326]).
  • clips : Boolean, if set true, clips the detected box into the image.