文档

FSRDetectionOutput

Input

  • rois:
    • Required: yes
    • Type: float16, float32
    • Description:

      Output proposal information, with shape [batch, 5-tuple, max_rois_num]. The 5-tuple is formatted as (batchID, x1, y1, x2, y2). max_rois_num is the maximum number of ROIs per batch (actual_rois_num applies) and must be a multiple of 16.

    • Restriction: Must be of type float16.
  • bbox_delta:
    • Required: yes
    • Type: float16, float32
    • Description: tensor of BBox delta, with shape [total_rois, num_classes * 4-tuple], where the 4-tuple is formatted as (delta_x, delta_y, delta_w, delta_h), and total_rois indicates the actual ROI count.
    • Restriction: Must be of type float16.
  • score:
    • Required: yes
    • Type: float16, float32
    • Description: tensor for the probability of each class, with shape [total_rois, num_classes]. Class 0 indicates the background probability and total_rois indicates the actual ROI count.
    • Restriction: Must be of type float16.
  • im_info:
    • Required: no
    • Type: float16, float32
    • Description: height and width of the input image
    • Restrictions: none
  • actual_rois_num:
    • Required: no
    • Type: int
    • Description: actual number of ROIs output per batch. Has shape [batch_rois, 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 ROIs per batch.

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

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

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

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

    • Restrictions: none

Attribute

  • batch_rois:
    • Required: no
    • Type: int
    • Description: batch size (default = 1). (This parameter is reserved and the value is obtained from the shape.)
    • Restrictions: none
  • num_classes:
    • Required: yes
    • Type: int
    • Description: number of classes, including the background
    • Restrictions: none
  • score_threshold:
    • Required: yes
    • Type: float
    • Description: score threshold
    • Restrictions: none
  • iou_threshold:
    • Required: yes
    • Type: float
    • Description: Intersection over Union threshold
    • Restrictions: none

Output

  • actual_bbox_num:
    • Required: yes
    • Type: int32
    • Description: actual number of output BBoxes, with shape [batch, num_classes], of type int32
    • Restrictions: none
  • box:
    • Required: yes
    • Type: float16, float32
    • Description: output proposal information, with shape [batch, numBoxes, 8-tuple], where the 8-tuple is [x1, y1, x2, y2, score, label, batchID, NULL], and numBoxes is up to 1024, that is, min(Maximum number of input boxes, 1024)
    • Restriction: Must be of type float16.