Documentation

PriorBox

Input

  • x:
    • Required: yes
    • Type: float16, float32
    • Description: input tensor. Only the height and width dimensions in the shape are used.
    • Restriction: Must be of type float16.
  • img:
    • Required: no
    • Type: float16, float32
    • Description: image information. Only the height and width dimensions in the shape are used. img_size or img_h/img_w in attr does not need to be specified.
    • Restriction: Must be of type float16.

Attribute

  • min_size:
    • Required: yes
    • Type: ListFloat
    • Description: minimum box size (in pixels)
    • Restrictions: none
  • max_size:
    • Required: yes
    • Type: ListFloat
    • Description: maximum box size (in pixels)
    • Restrictions: none
  • aspect_ratio:
    • Required: yes
    • Type: ListFloat
    • Description: ratio of the width and height of the generated box to the input box. Duplicate values are deleted.
    • Restriction: Must be of type listFloat. Must be greater than 0.
  • flip:
    • Required: no
    • Type: bool
    • Description: whether to flip each aspect_ratio (default = True)
    • Restrictions: none
  • clip:
    • Required: no
    • Type: bool
    • Description: whether to clip the generated boxes (default = False)
    • Restrictions: none
  • variance:
    • Required: no
    • Type: ListFloat
    • Description: variance of the generated boxes
    • Restrictions: none
  • step_h:
    • Required: no
    • Type: float
    • Description: stride height
    • Restrictions: none
  • step_w:
    • Required: no
    • Type: float
    • Description: stride width
    • Restrictions: none
  • offset:
    • Required: no
    • Type: float
    • Description: offset of each small grid relative to the upper left corner
    • Restrictions: none
  • img_h:
    • Required: no
    • Type: int
    • Description: image height
    • Restrictions: none
  • img_w:
    • Required: no
    • Type: int
    • Description: image width
    • Restrictions: none

Output

y:
  • Required: yes
  • Type: float16, float32
  • Description: output tensor, with shape [1, 2, num_output * 4], where num_output=H *W*num_priors, num_priors indicates the number of ROIs in each grid, calculated as: Number of min_size + Number of aspect_ratio * Number of min_size + Number of max_size.
  • Restriction: Must be of type float16.