文档

PointPillarDecodeBox

Input

x:

  • Required: yes
  • Type: float16
  • Description: trunk outputs of the PointPillar network
  • Restriction: NCHW. Multiple batches and quantized inference are not supported. align16(W) <= 10 x 1024

Attribute

  • box1_size:
    • Required: yes
    • Type: float16
    • Description: size of the first box, including the length, width, and height
  • box2_size:
    • Required: yes
    • Type: float16
    • Description: size of the second box, including the length, width, and height
  • box3_size:
    • Required: yes
    • Type: float16
    • Description: size of the third box, including the length, width, and height
  • box4_size:
    • Required: yes
    • Type: float16
    • Description: size of the fourth box, including the length, width, and height
  • rotations:
    • Required: yes
    • Type: float16
  • x_strides:
    • Required: yes
    • Type: float16
    • Description: stride of the X coordinate
  • y_strides:
    • Required: yes
    • Type: float16
    • Description: stride of the Y coordinate
  • z_strides:
    • Required: yes
    • Type: float16
    • Description: the start position of the box height
  • x_start:
    • Required: yes
    • Type: float16
    • Description: the start position of the X coordinate
  • y_start:
    • Required: yes
    • Type: float16
    • Description: the start position of the Y coordinate

Output

decode_boxes:
  • Required: yes
  • Type: float16
  • Description: output of the parsing box
  • Restriction: NCHW