Documentation

SSDDetectionOutput

Input

  • bbox_delta:
    • Required: yes
    • Type: float16, float32
    • Description: box offset, with shape [batch, N * num_loc_classes * 4]

      num_loc_classes_= share_location==True? 1: num_classes. The 4 is formatted as (xmin, ymin, xmax, yman).

    • Restriction: Must be of type float16.
  • score:
    • Required: yes
    • Type: float16, float32
    • Description: confidence value, with shape [batch, N * Num_classes]
    • Restriction: Must be of type float16.
  • anchors:
    • Required: yes
    • Type: float16, float32
    • Description: preselected box position formatted as [batch, 2, N * 4] or [1, 2, N * 4]. 2 indicates box and variance respectively. If variance_encoded_in_target==True, variance is not required (that is, 2 can be changed to 1). 4 formatted as (xmin, ymin, xmax, yman)
    • Restriction: Must be of type float16.

Attribute

  • num_classes:
    • Required: yes
    • Type: int
    • Description: number of classes to be predicted. If background_label_id==–1, this value must be greater than or equal to 1. If background_label_id >= 0, this value must be greater than or equal to 2. The maximum value is 1024, and the default value is 2.
    • Restriction: Must be within 1024.
  • share_location:
    • Required: no
    • Type: bool
    • Description: If True (default), classifies all classes as a category for position prediction.
    • Restrictions: none
  • background_label_id:
    • Required: no
    • Type: int
    • Description: background label ID (default = 0). Must be greater than or equal to –1.
    • Restriction: Must be greater than or equal to –1.
  • iou_threshold:
    • Required: no
    • Type: float
    • Description: Intersection over Union threshold (default = 0.3). Must be in the range (0, 1].
    • Restriction: Must be in the range (0, 1].
  • top_k:
    • Required: no
    • Type: int
    • Description: number of BBoxes before the NMS operation (default = 200). Must be in the range (0, 1024].
    • Restriction: Must be in the range (0, 1024].
  • eta:
    • Required: no
    • Type: float
    • Description: NMS parameter (default = 1). Must be 1.
    • Restriction: Must be 1.
  • variance_encoded_in_target:
    • Required: no
    • Type: bool
    • Description: If False, the result without using variance is brought into the position prediction.
    • Restriction:
  • code_type:
    • Required: no
    • Type: int
    • Description: BBox coding type (default = 1). 1: corner; 2: center_size; 3: corner_size
    • Restriction: Selected from 1, 2, and 3.
  • keep_top_k:
    • Required: no
    • Type: int
    • Description: number of BBoxes after the NMS operation (default = 200). Must be in the range (0, 1024]. If –1, keeps all BBoxes after the NMS operation.
    • Restriction: Must be in the range (0, 1024].
  • confidence_threshold:
    • Required: no
    • Type: float
    • Description: Confidence threshold. Must be in the range [0, 1].
    • Restriction: Must be in the range [0, 1].

Output

  • out_boxnum:
    • Required: yes
    • Type: int32
    • Description: maximum number of boxes output per batch. Has shape [batch, 8]. 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 boxes output per batch.

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

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

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

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

    • Restrictions: none
  • y:
    • Required: yes
    • Type: float16, float32
    • Description: output box data, with shape [batch, len, 8-tuple]. The 8-tuple is formatted as (batchID, label(classID), score (category probability), xmin, ymin, xmax, yman, null), where len is the value of keep_top_k aligned to 128. For example, if batch = 2 and keep_top_k = 200, the output shape is (2, 256, 8). The first 256 x 8 pieces of data are the result of the first batch.
    • Restriction: Must be of type float16.