Yolo
Input
x:
- Required: yes
- Type: float16, float32
- Description: Has shape [batch, boxes * (coords+1+classes), height, width].
- Restriction: Must be of type float16.
Attribute
- 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.
- classes:
- Required: no
- Type: int
- Description: number of classes (default = 80), up to 1024.
- Restriction: Must be within 1024.
- yolo_version:
- Required: no
- Type: string
- Description: Yolo version, either V2 or V3 (default), corresponding to detection_layer and Yolo_layer
- Restriction: V2 or V3
- softmax:
- Required: no
- Type: bool
- Description: whether to apply Softmax in YOLO V2 or V3 (default = False)
- Restriction: either True or False
- softmaxtree:
- Required: no
- Type: bool
- Description: whether to apply Softmaxtree (default = False). This attribute is reserved (see the definition in DarkNet).
- Restriction: False
- background:
- Required: no
- Type: bool
- Description: whether to work with softmax in YOLO V2 or V3 to determine the operation type for b and classes (default = False)
- Restriction: either True or False
Output
- coord_data:
- Required: yes
- Type: float16, float32
- Description: predicted coordinates, with shape [batch, boxes * coords, ceil(height * width * 2+32, 32)/2]
- Restriction: Must be of type float16.
- obj_data:
- Required: yes
- Type: float16, float32
- Description: [batch, ceil(boxes * height * width * 2+32, 32)/2], where each anchor has only one obj value
- Restriction: Must be of type float16.
- classes_data:
- Required: yes
- Type: float16, float32
- Description: [batch, classes, ceil(boxes * height * width * 2+32, 32)/2]. The score of each anchor is rounded up to the nearest multiple of 16 for calculation convenience.
- Restriction: Must be of type float16.
Parent topic: Supported Caffe Operators