YoloV3DetectionOutput

Input

• coord_data_low:
  • Required: yes
  • Type: float16, float32
  • Description: predicted coordinates, with shape [batch, coords *boxes, ceil(height_low*width_low,16) + 16]
  • Restriction: Must be of type float16 and height_low * width_low * Dtype_Size >= 32 bytes.
• coord_data_mid:
  • Required: yes
  • Type: float16, float32
  • Description: predicted coordinates, with shape [batch, coords * boxes, ceil(height_mid * width_mid,16) + 16]
  • Restriction: Must be of type float16 and height_mid * width_mid * Dtype_Size >= 32 bytes.
• coord_data_high:
  • Required: yes
  • Type: float16, float32
  • Description: predicted coordinates, with shape [batch, coords * boxes, ceil(height_high * width_high,16) + 16]
  • Restriction: Must be of type float16 and height_high * width_high * Dtype_Size >= 32 bytes.
• obj_prob_low:
  • Required: yes
  • Type: float16, float32
  • Description:[batch, ceil(boxes * height_low * width_low,16) + 16], where each anchor has only one obj value
  • Restriction: Must be of type float16.
• obj_prob_mid:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, ceil(boxes * height_mid * width_mid,16) + 16], where each anchor has only one obj value
  • Restriction: Must be of type float16.
• obj_prob_high:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, ceil(boxes * height_high * width_high, 16) + 16], where each anchor has only one obj value
  • Restriction: Must be of type float16.
• classes_prob_low:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, classes, ceil(boxes * height_low * width_low, 16) + 16]. The score of each anchor is rounded up to the nearest multiple of 16 for calculation convenience.
  • Restriction: Must be of type float16 and height_low * width_low * Dtype_Size >= 32 bytes.
• classes_prob_mid:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, classes, ceil(boxes * height_mid * width_mid, 16) + 16]. The score of each anchor is rounded up to the nearest multiple of 16 for calculation convenience.
  • Restriction: Must be of type float16 and height_low * width_low * Dtype_Size >= 32 bytes.
• classes_prob_high:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, classes, ceil(boxes * height_high * width_high, 16) + 16]. The score of each anchor is rounded up to the nearest multiple of 16 for calculation convenience.
  • Restriction: Must be of type float16 and height_low * width_low * Dtype_Size >= 32 bytes.
• img_info:
  • Required: yes
  • Type: float16, float32
  • Description: original image information, with shape [batch, 4-tuple], where the 4-tuple is formatted [netH, netW, scaleH, scaleW]. netH and netW are H and W of the network model input, and scaleH and scaleW are H and W of the original image.
  • Restriction: Must be of type float16.

Attribute

• biases_low:
  • Required: yes
  • Type: ListFloat
  • Description: [boxes, 2-tuple], where the 2-tuple is x (w) and y (h), corresponding to box 1
  • Restrictions: none

• biases_mid:
  • Required: yes
  • Type: ListFloat
  • Description: [boxes, 2-tuple], where the 2-tuple is x (w) and y (h), corresponding to box 2
  • Restrictions: none

• biases_high:
  • Required: yes
  • Type: ListFloat
  • Description: [boxes, 2-tuple], where the 2-tuple is x (w) and y (h), corresponding to box 3
  • Restrictions: none

• boxes:
  • Required: no
  • Type: int
  • Description: number of anchor boxes in each grid (default = 3)
  • Restrictions: none

• coords:
  • Required: no
  • Type: int
  • Description: number of coordinates. Fixed at 4, indicating x, y, h, and w.
  • Restriction: Must be 4.

Output

• box_out:
  • Required: yes
  • Type: float16, float32
  • Description: [batch, 6-tuple, post_nms_topn], where the 6-tuple is [x1, y1, x2, y2, score, label(class)]. box_out_num applies.
  • Restriction: Must be of type float16.

• box_out_num:
  • Required: yes
  • Type: int32
  • Description: number of valid boxes per batch, up to 1024. Has shape [batch, 8, 1, 1]. The value 8 indicates that only the first column in the eight columns is valid (data padding is performed for performance acceleration, and the rest seven columns are invalid data that is padded).

    For example, for shape [4, 8], only the first column is valid, indicating the number of valid boxes per batch.

    [0,0]...[0,7]

    [1,0]...[1,7]

    [2,0]...[2,7]

    [3,0]...[3,7]

  • Restriction: Must be of type float16.